Manuscript M1 15 10 2021

Master General Physics
Major Particles, Nuclei and Universe
Particles and Symmetries
Samuel Wallon
version: 15 October 2021
Contents
I. Relativistic symmetries 1
1. Lorentz group 3
1.1. Space-time interval . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2. Poincaré and Lorentz groups . . . . . . . . . . . . . . . . . . . 6
1.2.1. Characterization of P . . . . . . . . . . . . . . . . . . . 6
1.2.2. Structure of the Lorentz group . . . . . . . . . . . . . . 9
1.2.3. The two types of Lorentz transformations . . . . . . . . 11
1.2.4. Number of parameters . . . . . . . . . . . . . . . . . . . 13
1.3. Covariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.1. Contravariant and covariant vectors . . . . . . . . . . . 14
1.3.2. Covariance and contravariance applied to Minkowski space
17
1.3.3. Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3.4. A few examples . . . . . . . . . . . . . . . . . . . . . . 22
2. The Klein-Gordon equation 25

2.1. Reminder: nonrelativistic quantum mechanics . . . . . . . . . . 25
2.1.1. Correspondence principle and Schrödinger equation . . . 25
2.1.2. Probability current . . . . . . . . . . . . . . . . . . . . . 25
2.2. The Klein-Gordon equation . . . . . . . . . . . . . . . . . . . . 27
2.2.1. Looking for a linear relativistic equation . . . . . . . . . 27
2.2.2. Second order differential equation . . . . . . . . . . . . 27
2.2.3. Non relativistic limit . . . . . . . . . . . . . . . . . . . . 28
2.3. Klein-Gordon equation and scalar field action . . . . . . . . . . 28
2.3.1. Lagrangian formulation . . . . . . . . . . . . . . . . . . 29
2.3.2. Euler-Lagrange equations . . . . . . . . . . . . . . . . . 29
2.3.3. Application: field equations of a massive scalar field the-
ory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4. Physical content . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.1. Probability density and current . . . . . . . . . . . . . . 31
2.4.2. Non relativistic limit . . . . . . . . . . . . . . . . . . . . 33
2.4.3. Energy spectrum and probability density . . . . . . . . 33
1
Contents
2.4.4. Charge current and reinterpretation of the negative en-

ergy solution . . . . . . . . . . . . . . . . . . . . . . . . 34
2.5. In-depth study: global symmetries and Noether theorem . . . . 35
2.5.1. Variation of the action with respect to boundary config-
urations . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5.2. Noether theorem . . . . . . . . . . . . . . . . . . . . . . 36
2.5.3. Charge current . . . . . . . . . . . . . . . . . . . . . . . 37
II. Introduction to Quantum Electrodynamics 39

3. From nonrelativistic perturbation theory to Feynman diagrams 41
3.1. Time dependent perturbation theory . . . . . . . . . . . . . . . 41
3.1.1. Higher orders . . . . . . . . . . . . . . . . . . . . . . . . 47
4. Electrodynamics 51
4.1. Maxwell equations . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1.1. Local form . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1.2. Integral form . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2. Covariant formulation of electrodynamics . . . . . . . . . . . . 52
4.2.1. Field strength tensor and its dual . . . . . . . . . . . . . 52
4.2.2. Lorentz transformations . . . . . . . . . . . . . . . . . . 54
4.2.3. Relativistic invariants . . . . . . . . . . . . . . . . . . . 57
4.3. Covariant form of Maxwell’s equations . . . . . . . . . . . . . . 58
4.3.1. Maxwell’s equations . . . . . . . . . . . . . . . . . . . . 58
4.3.2. Four-potential . . . . . . . . . . . . . . . . . . . . . . . 60
4.4. Lagrangian for photons . . . . . . . . . . . . . . . . . . . . . . 63
4.5. Coupling between matter and electromagnetic field . . . . . . . 66
4.5.1. Matter part . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.5.2. Pure photon part . . . . . . . . . . . . . . . . . . . . . . 67
4.5.3. Interaction between matter and photons . . . . . . . . . 67
5. From Lagrangians to cross-sections 75

5.1. Perturbation theory . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2. Cross-sections . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.2.1. normalization of a free particle wave-function . . . . . . 78
5.2.2. Transition rate per unit volume . . . . . . . . . . . . . . 78
5.2.3. Number of final states . . . . . . . . . . . . . . . . . . . 79
5.2.4. Initial flux . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2.5. Cross-section . . . . . . . . . . . . . . . . . . . . . . . . 80
5.2.6. Cross-section in the center-of-mass system . . . . . . . . 81
2
Contents
Appendices 82
A. Elements of group theory 85
A.1. Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
A.1.1. Definitions . . . . . . . . . . . . . . . . . . . . . . . . . 85
A.1.2. Examples of groups . . . . . . . . . . . . . . . . . . . . 86
A.2. Morphisms and subgroups . . . . . . . . . . . . . . . . . . . . . 88
A.2.1. Maps, injections, surjections . . . . . . . . . . . . . . . 88
A.2.2. Morphism . . . . . . . . . . . . . . . . . . . . . . . . . . 89
A.2.3. Subgroup . . . . . . . . . . . . . . . . . . . . . . . . . . 90
B. Levi-Civita symbols and tensors 93

B.1. Levi-Civita symbol in an euclidean space . . . . . . . . . . . . . 93
B.1.1. 2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
B.1.2. 3d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
B.1.3. General case . . . . . . . . . . . . . . . . . . . . . . . . 94
B.1.4. Transformation under O(n) and SO(n) . . . . . . . . . 95
B.2. Levi-Civita tensor in 4d Minkowski space . . . . . . . . . . . . 96
B.2.1. Contraction identities . . . . . . . . . . . . . . . . . . . 96
B.2.2. Relationship with the Levi-Civita symbol . . . . . . . . 97
B.2.3. Behavior under Lorentz transformations . . . . . . . . . 97
3
Part I.
Relativistic symmetries
1
1. Lorentz group
Maxwell equations, which describe electric and magnetic phenomena, are not in-
variant under the Galilean group. Thus, contrarily to what states the galilean
principle, one should be able to reveal an absolute reference system, named
ether. Several experiments led by Michelson and Morley (1881) invalidated
such an hypothesis: speed of light is independent of the direction of propaga-
tion. Time cannot be considered as an absolute variable: simultaneity of two
phenomena in reference frame does not imply simultaneity in another one, in
contrast with classical mechanics.
1.1. Space-time interval

Let us briefly remind here the notion of space-time interval in special relativity.
First consider an event described by an observer, assumed to be fixed in an
inertial frame, by its space-time coordinate (ct1 , ~x1).
Example : Emission of a photon in π 0 → γγ disintegration:
The signal propagates at the speed of light (photon). It is detected at the

space-time coordinates (ct2 , ~x2).
By definition, the square of the space-time interval between two events is given
by
(∆s)2 = c2 (t2 − t1 )2 − (~x2 − ~x1)2 . (1.1)
Of course in the present case, ~x2 − ~x1 = c(t2 − t1 )~u where ~u is a unitary vector
pointing in the direction of the photon trajectory, and thus (∆s)2 = 0 .
A key assumption, made by Einstein, is the fact that c does not depend on the
inertial frame, therefore (∆s)2 = 0 in every inertial frame in which one can
wish to describe this experiment.
From the definition (1.1), the infinitesimal interval square reads
ds2 = c2 dt2 − dx2 − dy 2 − dz 2 . (1.2)
3
1. Lorentz group
If ds2 = 0 in an inertial frame K, then ds′2 = 0 in any other inertial frame K ′.

In the infinitesimal limit, one should thus have ds2 = a ds′ 2 .
Using the homogeneity of time and space, and isotropy of space, one can show
that a = constant = 1.
Exercise 1.1
Show this (see Landau, Lifshitz, The Classical Theory of Fields, p. 12).
The space of points in space-time with the interval (1.2) is called Minkowski
space1. This is called a pseudo-euclidean geometry, in contradistinction with
usual euclidean geometry.
A point in Minkowski space M is described by its contravariant coordinates
µ
x = (ct, ~x) . This is an example of a quadrivector.
Metric: it is given by the matrix

 
1
 −1 
gµν = . (1.3)
 −1 
−1
and thus
(∆s)2 = gµν (x2µ − x1µ ) (x2ν − x1ν ) ,
which is identical to the square of interval defined previously.
One can now define 3 kinds of intervals depending on the sign of (∆s)2 :
⋄ (∆s)2 > 0 time-like interval
⋄ (∆s)2 < 0 space-like interval
⋄ (∆s)2 = 0 light-like interval ,
as illustrated in Fig. 1.1.
The infinitesimal square of the space-time interval thus reads
ds2 = gµν dxµdxν . (1.4)
1
Hermann Minkowski introduced this concept of physical space in 1907. Henri Poincaré introduced the same
idea in 1905 from a mathematical point of view, in connexion with the invariance of Maxwell equations
under change of inertial frame.
4
1.1. Space-time interval
ct > 0
(∆s)2 > 0
0
=
s) 2
(∆
(∆s)2 < 0
spatial
components
ct < 0
Figure 1.1.: The 3 different kinds of intervals.
Exercise 1.2
Show that the interval separating the emission and the detection of a particle
in any process is either time-like or light-like.
A trajectory is given by a curve X µ (s) where s is an arbitrary parametrization

of this trajectory. The infinitesimal proper time dτ is defined through
(cdτ )2 = (cdt)2 − (d~x)2 . (1.5)
Denoting the velocity as
d~x
~v = (1.6)
dt
we have
r
p ~v 2
cdτ = (cdt)2 − (~v)2 dt2 = cdt 1− , (1.7)
c2
i.e.
dτ
dt = q (1.8)
~v 2
1− c2
or equivalently
dt = γdτ with γ = √1 and β~ = ~vc . (1.9)

1−β 2
5
1. Lorentz group
From the invariance of ds2 under change of inertial frame, we deduce that the
proper time dτ does not depend on the choice of inertial frame: it is a Poincaré
scalar. We will characterize Poincaré transformations in the next section.
The relation (1.9) reveals a key feature. Indeed, suppose that a particle is at
rest in a given inertial frame, named proper frame. This is always possible if
its velocity differs from c, using a pure Lorentz boost along ~v, see next section.
Passing to another arbitrary inertial frame, of non zero velocity with respect to
the proper frame, we will have γ > 1, which implies the famous phenomena of
time dilation: dt > dτ : this means that time runs slower for a particle in any
inertial frame with a non zero velocity with respect to a proper frame.
Throughout the rest of the text, unless otherwise stated, we will use a set of
units in which c = 1 , and thus x0 = t, x1 = x, x2 = y, x3 = z .
1.2. Poincaré and Lorentz groups

Having postulated that the speed of light is independent of the inertial frame,
we have seen the important consequence that the infinitesimal interval ds2 is
identical in every inertial frame. We now want to determine the relativistic
transformations relating inertial frames. We will then study the group structure
of these transformations.
1.2.1. Characterization of P
Consider a point x of M, measured in frame K, and x′ the coordinates of the
same point in frame K ′.
The invariance of the infinitesimal interval
ds′ 2 = ds2 (1.10)
implies that
translation (4 real parameters)

ւ
x = Λ β x + aα : Poincaré group P (10 real parame-
′α α β
(1.11)
ters)
տ
Lorentz group L (6 real parameters)
Proof :
6
The condition (1.10) reads
gαβ dx′αdx′β = gµν dxµ dxν ,
i.e.
∂x′α ∂x′β µ ν
gαβ dx dx = gµν dxµdxν
∂xµ ∂xν
thus
∂x′α ∂x′β
gαβ = gµν . (1.12)
∂xµ ∂xν
Taking the determinant of each side of this equation, one thus gets
2
∂x′
det g det = det g .
∂x
∂x′
Since g is regular, this implies that det 6= 0 : Poincaré transformations are
∂x
thus invertible.
∂
Acting with on Eq. (1.12), one gets
∂xρ
∂ 2x′α ∂x′β ∂x′α ∂ 2x′β
gαβ + gαβ =0
∂xµ∂xρ ∂xν ∂xµ ∂xν ∂xρ
which symbolically reads
A(µρ)ν + A(νρ)µ = 0 , (1.13)
where A is symmetric with respect to indexes inside brackets.
µ ↔ ρ gives A(µρ)ν + A(µν)ρ = 0 , (1.14)
ν ↔ ρ gives A(µν)ρ + A(νρ)µ = 0 . (1.15)

The combination (1.13) + (1.14) -(1.15) thus gives 2A(µρ)ν = 0, i.e.
∂ 2x′α ∂x′β
gαβ =0
∂xµ∂xρ ∂xν
∂x′ ∂ 2x′α
and thus (since det 6= 0) : =0
∂x ∂xµ∂xρ
which proves that x′ are linear functions of x .
7
1. Lorentz group
We use matrix notations, namely
ւ line index
α
Λ β
տ column index
One deduces from Eqs. (1.12) and (1.11) that Λ is a real matrix satisfying
gαβ Λαµ Λβν = gµν . (1.16)
The set L of Lorentz transforms is a group (see Appendix A for a very short
introduction to group theory).
Proof :
• this is obvious from the definition: the product of two such homogeneous
transformations preserving ds2 is an homogeneous transformation which pre-
serves ds2 ; on the hand, the inverse of an homogeneous transformation pre-
serving ds2 is an homogeneous transformation preserving ds2 . And of course
the identity preserves ds2.
• algebraically, if Λ and Λ′ belong to L, then

gαβ Λαµ Λβν = gµν
gαβ Λ′ α µ Λ′ β ν = gµν .
This implies that
µ β ′ν µ ′ν
gαβ Λαµ Λ′ µ′ Λ ν Λ ν ′ = gµν Λ′ µ′ Λ ν ′ = g µ′ ν ′ ,
and thus that Λ Λ′ ∈ L .
The set P of Poincaré transformations, characterized by a pair (a, Λ) where

a is a 4-vector and Λ ∈ L, is a group.
Proof :
The first of the two above proofs immediately extends to the case of the inho-
mogeneous Lorentz group (or Poincaré group). The second proof relies on the
group law on P :
8
given two Poincaré transformations (a, Λ) and (a′ , Λ′ ), it is an easy exercise to

show that their product reads
(a, Λ) · (a′ , Λ′) = (Λ′ a + a′ , Λ′ Λ) (1.17)
of the required form (a′′ , Λ′′ ).
On should notice that the defining relation (1.16) of L reads, in matrix form,
Λt g Λ = g . (1.18)
This is easily recovered in the following way: Λ being a linear transformation,

which leaves the scalar product invariant, one has, for arbitrary quadrivectors
v and v ′ of M, denoted as column vectors,
v ′ · v = (v ′)t gv = (Λv ′)t g(Λv) = (v ′)t Λt gΛv (1.19)
which is equivalent to (1.18).
1.2.2. Structure of the Lorentz group

From Eq. (1.16) one deduces that det Λ = ±1 :

det Λ = +1 : proper transformations L+

det Λ = −1 : improper transformations L− .
2 P3 2
Again considering Eq. (1.16) with µ = ν = 0 , one gets Λ00 − Λi 0 = 1 :
i=1
0
Λ0>1 :
0 no time reversal (orthochronous) L↑
Λ 0 6 −1 : with time reversal (antichronous) L↓ .
One can show (see Gelfand p. 165) that
The Lorentz group O(3, 1) has 4 connected component. Each of

them is doubly connected, but not simply connected.
Connected component: in a connected component, one can pass from one ele-
ment to another one continuously.
Simply and multiply connected space:
A space is simply connected if any loop in this space can be contracted to a

point. Equivalently, considering two paths, one can pass continuously from the
9
1. Lorentz group
first one to the second one. If not, it is multiply connected. In our present
case, doubly connected means that in each of these components, there are two
classes of paths; considering two paths in a given class, one can pass from the
first one to the second one by continuous transformation of the parameters.
L↑+ L↑−
P
Λ00 > 1 Λ00 > 1
det Λ = 1 det Λ = −1
T T
PT
L↓− L↓+
P
Λ00 6 −1 Λ00 6 −1
det Λ = −1 det Λ = 1
Figure 1.2.: The four connected components of L.
These various connected components are related by discrete transformations:
• space inversion (or parity):

 
1
 −1 
P = 
 −1 
−1
• time reversal:  
−1
 1 
T = 
 1 
1
10
This is illustrated in Fig. 1.2.
L leaves the 3 regions x2 > 0, x2 = 0 (light-cone) and x2 < 0 invariant.

↑
x2 > 0 x2 > 0
L+ leaves the 4 regions , , x2 = 0 , x2 < 0 invariant, as
x0 > 0 x0 < 0
shown in Fig 1.1.
L↑+ is the restricted Lorentz group
L↑+ and L↑− form together the complete Lorentz group
L↑+ and L↓+ form together the proper Lorentz group
Exercise 1.3
Check that the restricted Lorentz group, the complete Lorentz group and the
proper Lorentz group are subgroups of the Lorentz group L .
Compactness:
The restricted Lorentz group L↑+ is non compact.
This is due to the existence of boosts, also called special Lorentz transforms,
which we will examine later, encoded by a rapidity which varies from −∞ to
+∞.
1.2.3. The two types of Lorentz transformations

We restrict ourselves to the restricted Lorentz group.
Rotations
The first set of Lorentz transformations is well known: they are the usual
rotations, elements of SO(3), defined as 3-dimensional transformations which
preserve the euclidean scalar product. They are characterized by the angle ψ
of the rotation, and the axis of the rotation, itself defined through a unitary
vector ~n depending on two spherical angles θ, φ. Thus, a rotation is defined by
three real numbers varying in compact sets
0 6 θ 6 π, 0 6 φ < 2π, 0 6 ψ 6 π , (1.20)
using the fact that a rotation around an oriented axis spanned by ~n and of
angle ψ is identical to a rotation of axis −~n and angle 2π − ψ, which allows to
11
1. Lorentz group
restrict ourselves to 0 6 ψ 6 π .
Exercise 1.4
Show that a rotation of axis x1 and angle Ψ in the passive point of view (i.e.
the axes rotates) reads
2′
x = cos ψ x2 + sin ψ x3
(1.21)
x3′ = − sin ψ x2 + cos ψ x3 .
Exercise 1.5
Show that a passive rotation of axis ~n (~n2 = 1) and angle Ψ reads
~x′ = (~x − ~n(~n · ~x)) cos θ − ~n ∧ ~x sin θ + ~n(~n · ~x) (1.22)
Indication: consider how each part of the decomposition of an arbitrary vector

~x as a sum of a component ~xk along the axis of the rotation and a component
~x⊥ perpendicular to it
~x = ~xk + ~x⊥ = [(~x · ~n)~n] + [~x − (~x · ~n)~n]
transforms under a rotation.
Boosts
Besides, pure Lorentz transformations, also named boosts, affect a single space
direction and time.
Exercise 1.6
Show that restricted Lorentz transforms which leaves invariant x2, x3 reads
0′
x = cosh φ x0 − sinh φ x1
(1.23)
x1′ = − sinh φ x0 + cosh φ x1
where φ ∈] − ∞, ∞[.
Indication: consider a generic 2 × 2 linear transformation mixing x0 and x1,
leaving the space-time interval invariant and preserving the sign of x0, and
solve for the 4 coefficients.
In the passive point of view, Λ encodes the transformation from a frame F to

a frame F ′ , while in the active point of view, it encodes the change of a point
12
in a fixed frame. In the passive point of view, the movement of the origin O′ of
the new frame F ′ with respect to the origin O of the initial frame F is obtained
by setting x1′ = 0, i.e. x1 = tanh φ x0 corresponding to a speed ~v = c tanh φ e~1
of the frame F ′ with respect to the frame F , along the x axis.
One should note that introducing β(−1 6 β 6 1) as

~v = cβ e~1 , (1.24)
and γ (0 6 γ) as
1
γ=p , (1.25)
1 − β2
we have
β = tanh φ , (1.26)
1
γ = p = cosh φ , (1.27)
1 − tanh2 φ
γβ = sinh φ , (1.28)
so that the pure boost (1.23) can be equivalently rewritten, β being algebraic,
0′
x = γ x0 − γβ x1
(1.29)
x1′ = −γβ x0 + γ x1
Exercise 1.7
Show that a boost in the direction ~n (~n2 = 1), of rapidity φ, i.e. with a speed
~v = c thφ ~n of the frame F ′ with respect to the frame F , reads
0′
x = x0 cosh φ − (~n · ~x) sinh φ
′ (1.30)
~x = ~x − ~n (~x · ~n) − x0 sinh φ + (~n · ~x) cosh φ ~n
Indication: consider how each part of the decomposition ~x = [(~x · ~n)~n] + [~x −
(~x · ~n)~n] transforms under a boost.
Generally speaking, a boost is therefore characterized by three real numbers

φ1 , φ2, φ3 .
1.2.4. Number of parameters

One can demonstrate that any element Λ of the restricted Lorentz group can
be written in the form
Λ = RB (1.31)
13
1. Lorentz group
where R is a rotation and B is a boost. We thus deduce that
L depends on 6 real parameters. (1.32)

and
P depends on 10 real parameters: (1.33)
it depends on 6 real parameters describing L to which one should add the 4
space-time translations.
1.3. Covariance
1.3.1. Contravariant and covariant vectors
In a formal way, a vector belongs to a vector space E (over the field K=R or
C), spanned by a basis. This naturally gives a set of contravariant coordinates
for this vector on this basis. A covector belongs to the dual space E ∗ of linear
forms2 E → K. In the special case of Hilbert spaces, there is a one-to-one
correspondence between E and E ∗, which is extremely convenient in practice.
Let us explain these different steps in detail.
Contravariance
Consider a vector space E (over the field K=R or C), of dimension n. A
′ ′
change of basis from B = {~e1 , · · · ~en } to B ′ = {~e1 , · · · ~en } (conveniently, basis
are denoted as row vectors) is given by
B ′ = BA (1.34)
where A is an invertible n × n matrix, with entries Ai j . This reads explicitly3

′
~ej = ~ei Ai j . (1.35)
A vector ~v can be expanded in the basis B, with notations which emphasize

the fact that the basis B is used,
~v = v i[B]~ei = B v[B] , (1.36)
2
We recall that a linear form is a linear map from a vector space to its field of scalars .
3
If not stated explicitly, we will now use Einstein convention, which states that repeated indexes are summed.
14
1.3. Covariance
where the column vector of components v i[B] is denoted as v[B],

 1 
v [B]
 v 2[B] 
v[B] = 
 ... 
 (1.37)
v n[B]
which allows the use of a matrix product in the second equality of Eq. (1.36).
In the new basis B ′ , the vector ~v being invariant, one has the two equivalent
expansions
~v = B v[B] = B ′ v[B ′ ] = BA v[BA] (1.38)
and thus
v[BA] = A−1v[B] (1.39)
Such a transformation law is named contravariant, since one has under the
transformation A
B −→ B ′ (1.40)
v[B] ←− v[B ′ ] . (1.41)
Covariance
Consider now the dual space E ∗ , the space of linear forms from E to K.
A natural (thus named canonical) dual basis can be defined as a set of n
linear forms ei , through their action on the chosen basis of E, as
ei (~ej ) = δji . (1.42)
Any linear form ℓ ∈ E ∗ can then be expanded in this basis as
ℓ = ℓi[B]ei = ℓ[~ei ]ei . (1.43)
Indeed any coefficient ℓi [B] of this expansion can be easily obtained through
the action of ℓ on a vector basis ~ei :
ℓ(~ei ) = ℓj [B]ej (~ei) = ℓj [B]δji = ℓi [B] . (1.44)
This expansion can be written in a more compact and useful form, as
ℓ = ℓi [B]ei = ℓ[~ei ]ei = ℓ[B]B ∗ , (1.45)
in which the row vector of components ℓi [B] is denoted as
ℓ[B] = (ℓ1[B], · · · , ℓn[B]) = (ℓ(~e1 ), · · · , ℓ(~en)) (1.46)
15
1. Lorentz group
and a column vector of components ei is denoted as B ∗

 1 
e
∗
 e2 
B =  ... 
 (1.47)
en
which allows the use of a matrix product in the last equality of Eq. (1.45).
The action of ℓ ∈ E ∗ on an arbitrary vector ~v ∈ E thus reads
ℓ(~v ) = ℓ(~ei)ei (~v) = ℓ(~ei )ei (v j~ej ) = ℓ(~ei )v j ei (~ej ) = ℓ(~ei )v j δji = ℓ(~ei )v i , (1.48)
were we have used the expansion of ~v on the basis B, see Eq. (1.36). Again
using our compact notations (row vectors for forms, column vectors for vectors),
we can write this as
ℓ(~v ) = ℓ(~ei)v i = ℓ[B]v[B] . (1.49)
Let us now investigate how the change of basis (1.34) for E translates in the
dual space. We have
ℓi[B ′ ] = ℓj [B]Aj i . (1.50)
Indeed, from Eqs. (1.44) and (1.35)
ℓi [B ′] = ℓi [BA] = ℓ(~ei ) = ℓ(~ej Aj i ) = ℓ(~ej )Aj i = ℓj [B]Aji .

′
(1.51)
Written in a more compact form, this means that
ℓ[B ′] = ℓ[B]A (1.52)
Such a transformation law is named covariant, since one has under the trans-
formation A
B −→ B ′ (1.53)
ℓ[B] −→ ℓ[B ′ ] . (1.54)
Forms are therefore also named co-vectors.
Correspondence between vector and co-vectors in the case of a Hilbert

space
In the case of a Hilbert space (in finite dimension, a vector space equipped with
a scalar product is automatically a Hilbert space; this is not always true for
16
1.3. Covariance
infinite dimension spaces), the Riesz representation theorem ensure that there
is a one to one correspondence between linear forms and vectors:
Denoting the metric as g, and ( , )g the associated scalar product,
∀ℓ ∈ E ∗, ∃! ~ℓg ∈ E / ∀~v ∈ E, ℓ(~v ) = (~ℓg , ~v)g . (1.55)
In practice, the scalar product is usually denoted without explicit reference to

the underlying metric, so that the index g is just removed.
The above discussion is familiar in quantum mechanics, where a form φ of
the dual space EH∗ of the Hilbert space EH is denoted as hφ| (bra) and means, by
the correspondence theorem, hφ| = (φ, ·)g or explicitly, when acting on a vector
|ψi (ket), hφ|ψi = (φ, ψ)g .
Coming back to Eq. (1.49), and removing any reference to g, we can now write
ℓ(~v ) = ℓi v i = (~ℓ, ~ei ) v i . (1.56)
1.3.2. Covariance and contravariance applied to Minkowski

space
Let us consider a space of dimension d, with a basis eµ (µ=1,··· ,d) , equipped with
a metric g. As a special case, the Minkowski M can be described through
a basis (e1, e2 , e3, e4) denoted eµ (µ=0,··· ,3) , the metric being given by Eq. (1.3).
Note that we avoid here using arrows, since they are traditionally used for the
spatial part of a 4-d vector.
Contra-variant components
An arbitrary quadrivector A can be decomposed in such a basis as
A = Aµ eµ . (1.57)
The Aµ are the contra-variant components of A.
Remark : the eµ are not vectors since they are not invariant when changing the
axis (recall that by definition a vector is an object with d components which is
invariant under changes of axis).
17
1. Lorentz group
Co-variant components
Using the metric g which allows through (1.55) to define a co-vector associated
to the vector A, the co-variant components Aµ of this co-vector reads, according
to Eq. (1.56),
Aµ = A · e µ (1.58)
In general Aµ 6= Aµ . As an example, consider the euclidean plane, see Fig. 1.3,
with a non orthonormal basis.
A2
A2 A
e2
e1 A1 A1
Figure 1.3.: Contravariant and covariant components.
Relation between co-variant and contra-variant components

By definition, the metric tensor (or metric) is related to the vector basis as
eµ · eν = gµν
with gµν given by (1.3) in the case of Minkowski space. We thus have
Aµ = A · eµ = Aν eν · eµ = gµν Aν
We define g µν as the inverse of gµν : g µν gνρ = δρµ .
In the case of Minkowski space, since gµν is its own inverse, gµν = g µν .
Thus
Aµ = gµν Aν
(1.59)
Aµ = g µν Aν .
This implies that
A · B = Aµ B ν gµν = Aµ Bµ = Aµ B µ . (1.60)
18
1.3. Covariance
Back to space-time interval

The expression of the scalar-product (1.60) implies in particular that
ds2 = dx · dx = dxµ dxν eµ · eν = dxµdxν gµν ,
in agreement with the above given definition (1.4).
One can also rely on co-variant space-time interval, so that we have the
following identities
ds2 = gµν dxµdxν = dxµ dxµ = g µν dxµdxν . (1.61)
Inverse of a Lorentz transformation in covariant notations

Let us consider a transformation Λ of L. From
gαβ Λαµ Λβν = gµν ,
one gets ν
gαβ Λαµ = gµν Λ−1 β
thus ν
Λ−1 β
= gβα Λαµ g µν .
We denote
Λβν = gβα Λαµ g µν . (1.62)
Thus
ν
Λ−1 β
= Λβν . (1.63)
Eq. (1.62) means that Λαµ is a tensor of order 2 which transforms as Aα Aµ , see
section 1.3.3. ν β
We thus have Λβν = Λ−1 β = t Λ−1 ν and one can check explicitly that
ν ν
Λβν Λρν = Λ−1 β Λρν = Λρν Λ−1 β = δβρ .
Remark :
With these notations, since 1 = 1−1 = 1t , one can write

δ µν = δν µ = δ ν µ = δµ ν = δνµ .
Conclusion:
The inverse of a matrix of L is easily obtained by rising and lowering indexes

using the metric tensor, and then transpose.
19
1. Lorentz group
Transformation law of co-variant components under L

Under a transformation Λ of L, the contra-variant components transform as
x′µ = Λµν xν (1.64)
according to Eq. (1.11). This implies that
x′µ = Λµν xν (1.65)

Proof :
This is just paste and copy of the relation (1.50) with here A = Λ−1 :
x′µ = xν (Λ−1)νµ = Λµν xν . (1.66)
More directly, one has indeed
x = xµeµ = x′ν e′ν = Λνµ xµe′ν (1.67)
and thus
eµ = Λνµ e′ν and e′µ = Λµν eν
according to the tensor notation introduced for Λ−1. Thus x′µ = x · e′µ =
x · Λµν eν = Λµν xν .
Remark :
The fact that covariant components transform under t Λ−1 , which is the inverse
transpose of the matrix encoding the transformation of contra-variant compo-
nents is due to the invariance of ds2 = dxµdxµ which implies that Λt gΛ = g,
see Eq. (1.18).
Indeed, this also reads t Λ−1 = g Λ g −1 . Starting from v ′ = Λ v for a con-
travariant v, one has
g v ′ = g Λ g −1 g v
|{z} | {z } |{z}
t −1
covariant vector Λ covariant vector
1.3.3. Tensors
Based on the tools introduced above, we can introduce relativistic covariance
in a more general context: physical quantities describing the state of a system
20
1.3. Covariance
at generally speaking not the same for different observers. We can therefore
distinguish:
• scalar quantities, which do not depend on a specific choice of reference
frame. These are defined by scalar functions: given a function φ defined on M,
in a given frame F , the corresponding scalar function in a new frame F ′ is such
that φ′ (x′) = φ(x) in order to leave invariant the associated physical quantity.
• the contravariant components v µ of a vector transform as coordinates (in
the general case of P, translations only affect coordinates) :
′µ

∂x
v ′µ = Λµν v ν = vν .
∂xν
• the covariant components vµ of a vector transform as xµ :

vµ′ = Λµν xν .
• tensor of rank (|{z}
m , |{z}
n )
co- contra-
µ ···µn
F ′ α1 ···αm 1 = Λα1β1 · · · Λαm βm Λµ1ρ1 · · · Λµn ρn Fβ1 ···βmρ1 ···ρn .
• derivation :
∂φ ∂xν ∂φ ∂xν
= ′µ with Λµν = ′µ .
∂x′µ ∂x ∂xν ∂x
Indeed,
∂xν
x′µ = Λµν xν and xν = (Λ−1)νµ x′µ = Λµν x′µ so that Λµν = .
∂x′µ
∂φ
Thus transforms as a covariant vector, denoted as ∂ν φ .
∂xν
From the definition of contravariant coordinates xµ = (x0, ~x) one gets

∂ ∂ ~
∂µ = = ,∇ , (1.68)
∂xµ ∂x0

∂ ∂ ~ .
∂µ = = , −∇ (1.69)
∂xµ ∂x0
~ is covariant.
Beware of signs : ~x is contravariant but ∇
The quadri-divergence of a quadri-vector is the scalar
µ ∂A0 ~ ~
µ
∂ Aµ = ∂µ A = +∇·A
∂x0
21
1. Lorentz group
This leads naturally to the introduction of the scalar d’Alembertian operator:
∂2µ
⊔ = ∂µ ∂ = 2 − ∆ .
⊓ (1.70)
∂x0
Summary :
• one can pass from one to another form of a given tensor by raising or
lowering indexes through the metric tensor, as one does for quadri-vectors.
• Lorentz transformation should act by letting its representative matrix act-
ing on the left on a quadri-vector (either on its contravariant or covariant com-
ponents), summing over repeated indexes (one up, one down). For contravariant
components, this is just the matrix Λ itself, and for covariant components, its
transpose inverse.
1.3.4. A few examples

Let us consider a few examples of tensors.
Proper time
The first trivial example of Lorentz scalar is of course provided by the proper
infinitesimal time dτ of any particle: it is a Lorentz invariant (which equals zero
for a massless particle).
Four-velocity
It is defined has

µ dX µ dt d~x
V = = c , = (γc, γ~v) (1.71)
dτ dτ dτ
which satisfies
V 2 = γ 2(c2 − ~v 2) = c2 .
The above definition makes sense for a massive particle. For a massless particle,
the above discussion fails since dτ = 0 and γ → ∞. Still, in any reference frame,
we have d~x
dt = c~n where ~n is a unit vector pointing in the direction of the motion
of the particle, so that differentiating with respect to t instead of τ leads to
V µ = (c, c ~n) ,
which now satisfies V 2 = 0 .
22
1.3. Covariance
Four-momentum
For a massive particle, it is defined as

µ dX µ
µ E
P = mV = m = (γmc, γm~v) = , p~ (1.72)
dτ c
which satisfies
P 2 = m2 c2 .
The above definition makes sense for a massive particle. For a massless particle,
the last equality is still valid, so that for arbitrary mass, we define

E
Pµ = , p~ . (1.73)
c
Note that in any case,
~v ~p
= . (1.74)
c2 E
Four-current
From the usual charge density and the current density, one can define a four-
current as

µ ~
J (x) = cρ(x), j(x) . (1.75)
Introducing the rest charge density ρ0 (i.e. the charge density in a rest frame),
related to the charge density ρ = ρv in a frame moving at velocity ~v with respect
to the proper frame, one easily shows that
J µ (x) = (cρv (x), ρv (x)~v) = ρ0 V µ .
p
with4 ρv = γρ0, with as usual γ = 1/ 1 − v 2 /c2 , which shows that J µ are the
contravariant components of a four-vector (ρ0 is a Lorentz scalar).
Let us consider the special case of the four-current associated to the motion
~
of a charged particle, of charge q, moving along a trajectory X(t). We then
have
~
ρ(ct, ~x) = q δ (3) (~x − X(t)) (1.76)
~
~j(ct, ~x) = q dX δ (3) (~x − X(t))
~ . (1.77)
dt
3
4
Indeed ρ = dd3 xq and d3 xv = γ1 d3 xr when expressing the infinitesimal volume in the boosted frame (“v“) in
terms of the infinitesimal volume in the boosted frame (“r”), thus ρv = γρ0 .
23
1. Lorentz group
The trajectory reads

~ )) .
X µ (T ) = (cT, X(T (1.78)
It is therefore natural to write correspondingly, based on the idea that the

four-current “feels” the charge at point x if and only if it is there, covering any
possibility by integrating over the whole trajectory,
Z
dX µ
J µ (x) = qc dT (T ) δ (4) (x − X(T )) . (1.79)
dT
First, we note that this expression is invariant under reparameterization since
dT cancels in this expression. Furthermore, this expression satisfies covariance.
3
Indeed, d4 x is a scalar (under a boost, dx0 → γdx0 and d3 x → dγx ) and thus
δ (4) (x) is also a scalar, since
Z
d4x δ (4) (x) = 1 .
Thus, J µ (x) transforms as dX µ , i.e. as the contravariant components of a four-

vector.
Inserting (1.78) inside the current (1.79), and since
!
dX µ
dX~
(T ) = c, (T ) , (1.80)
dT dT
we get for the four-current

Z !
dX~
~
J µ (ct, ~x) = qc dT δ (3) (~x − X(t)) δ(ct − cT ) c, (T ) (1.81)
dT
!
d ~
X
~
= cqδ (3) (~x − X(t)), q ~
δ (3) (~x − X(t)) (1.82)
dt

= cρ(x), ~j(x) (1.83)
as expected.
24
2. The Klein-Gordon equation
In this chapter, our aim is to write a relativistic equation describing a massive
spin 0 particle. Meanwhile, we will see that such a description naturally leads
to states of negative energy, which will turn to be negative probability density
solutions. This will be reinterpreted as antiparticles.
2.1. Reminder: nonrelativistic quantum mechanics

2.1.1. Correspondence principle and Schrödinger equation
We first start with the way Schrödinger equation can be obtained from the
correspondence principle. Starting from the nonrelativistic dispersion equation
p2
E= (2.1)
2m
and making the replacement, according to the correspondence principle,
∂
E → i~ (2.2)
∂t
~~
~p → ∇ , (2.3)
i
we get
~2 ∂ψ
− ∆ψ(~x, t) = i~ , (2.4)
2m ∂t
which is the Schrödinger equation for a free particle.
2.1.2. Probability current

The probability density
ρ = |ψ|2 (2.5)
has the physical meaning that
d3 P = |ψ|2 d3 V (2.6)
25
is the probability of finding a particle in the elementary volume d3 V .

On the one hand, starting from the Schrödinger equation (2.4), multiplied by
∗
ψ , and on the other hand writing its complex conjugate, multiplied by ψ, we
get respectively
~2 ∗ ∂ψ
− ψ ∆ψ = i~ ψ ∗ (2.7)
2m ∂t
2
~ ∗ ∂ψ ∗
− ψ∆ψ = −i~ ψ (2.8)
2m ∂t
and by subtraction
~2 ∂ψ
− (ψ ∗ ∆ψ − ψ∆ψ ∗ ) = 2i~ ψ ∗ (2.9)
2m ∂t
so that
∂ρ ∂ψ i~
= 2ψ ∗ = (ψ ∗ ∆ψ − ψ∆ψ ∗ ) (2.10)
∂t ∂t 2m
~ 2,
i.e., since ∆ = ∇
∂ρ i~ ∗ ~ 2 ~ 2 ∗
i~ ~ ∗ ~ ~ ∗

= ψ ∇ ψ − ψ∇ ψ = ∇ · ψ ∇ψ − ψ ∇ψ (2.11)
∂t 2m 2m
which reads
∂ρ ~ ~
+∇·j =0 (2.12)
∂t
with the probability current, also named probability flux density, given by

~j = − i~ ψ ∗ ∇ψ
~ − ψ ∇ψ
~ ∗ . (2.13)
2m
Eq. (2.12) expresses the local conservation of probability. The global point of
view is obtained by integration over a given volume V : the increase of the
probability for particles to be inside V , of boundary S, is given by
ZZZ ZZZ ZZ
∂ 3 ~ · ~j d V = −
3 ~
~j · d2 S
ρd V = − ∇ (2.14)
∂t V V S
where we have used the Green-Ostrogradsky theorem in the last step. Eq. (2.14)
simply states that this increase is equal to the opposite of the flux of particles
out of the volume V , i.e. through S.
26
2.2. The Klein-Gordon equation
A plane wave solution of the Schrödinger equation (2.4)

i
ψ = N e ~ (~p·~x−Et) , (2.15)
which describes a free particle of momentum p~ and energy E has therefore a

probability density and a current
ρ = |N |2 , (2.16)
~j = p~ |N |2 = ~v |N |2 . (2.17)
m
2.2. The Klein-Gordon equation

2.2.1. Looking for a linear relativistic equation
The Schrödinger equation (2.4) explicitly violates Lorentz covariance. In special
relativity, we have
E2
p pµ = 2 − p~ 2 = m2 c2
µ
(2.18)
c
and thus
p
E= p~ 2c2 + m2 c4 . (2.19)
Looking for a first order equation, in the spirit of the Schrödinger equation, the
correspondence rule would thus lead to write
2

∂ψ p 2 2 2 ~
i~ = −~ c ∆ + m2 c4 ψ = mc 1 − ∆ + ··· ψ. (2.20)
∂t 2m2c2
This equation is very non-local, and the symmetry between ct and ~x is com-
pletely hidden.
2.2.2. Second order differential equation

Giving up with such first order equations1, and writing the four-momentum
operator as2

µ i~ ∂ ~ = i~ ∂ µ,
p = , −i~∇ (2.21)
c ∂t
1
Such an attempt makes sense when looking for a relativistic equation describing a spin 1/2 particle: this is
the essence of the construction of the Dirac equation historically.
2
Indeed ∇~ = ∂i = −∂ i .
27
the quadratic dispersion (2.18) leads, using the correspondence principle, to

2 µ
−~ ∂ ∂µ − m2 c2 Φ = 0 (2.22)
i.e.
2 2

m c
∂ µ ∂µ + 2 Φ(x) = 0 (2.23)
~
or equivalently, setting c = ~ = 1,

+ m2 Φ(x) = 0 , (2.24)
which is the Klein-Gordon equation.

This equation satisfies Lorentz covariance since under a transformation x′ =
Λx, denoting φ′ (x′) = φ(x), Eq. (2.24) becomes
′
+ m2 Φ′ (x′) = 0 , (2.25)
′
since ′ = ∂ µ ∂µ′ = ∂ µ ∂µ = , as any scalar operator.
2.2.3. Non relativistic limit

In order to study the classical (i.e. non relativistic) limit, let us extract the
rest-mass energy in the time-dependence of the wave function. We thus write
mc2
1 −i t
Φ(t, ~x) = √ e ~ Ψ(t, ~x) , (2.26)
2m
and thus

1 ∂2 2m ∂ ∂2
−i − Ψ(t, ~x) = 0 . (2.27)
c2 ∂t2 ~ ∂t ∂~x 2
In the limit c → ∞ , neglecting the first term, this equation simplifies into
∂ ~2
i~ =− ∆Ψ(t, ~x) , (2.28)
∂t 2m
which is the Schrödinger equation, as expected.
2.3. Klein-Gordon equation and scalar field action

As a discursive remark, let us show that the Klein-Gordon equation is the
equation of motion of a free scalar field.
28
2.3. Klein-Gordon equation and scalar field action
2.3.1. Lagrangian formulation

We consider the generic situation of a field theory, i.e. a system with an infinite
number of degrees of freedom, indexed by the space-time position3.
In any local field theory, the action reads
Z
S = d4 x L(x) (2.29)
where L(x) is the lagrangian density (usually called lagrangian), which is as-
sumed to depend on a finite number of derivatives of fields4. In practice, in
general L(x) only depends on fields and their first order derivatives, just like
in mechanics, the Lagrange function only depends on generalized coordinates
and their first order time derivative, so that the equations of motion would be
second order equations5 Thus, L(x) has the generic form
L(x) = L(φi (x), ∂n φi (x), x) . (2.30)
The index i labels the various types of fields or, if field are not scalars (this will
not be the case in the present lectures), their components (example: Aµ (x) in
the case of electromagnetism, or Ψ(x), a bispinor field (a collection of 4 complex
numbers) describing a spin 1/2 relativistic particle). Generally speaking, we will
always assume that fields vanish fast enough at infinity, so that boundary terms
can be safely ignored in usual integration by parts. Note that at that stage, the
space-time group of transformation could be classical (galilean) or relativistic
(lorentzian). For simplicity, we will now use covariant relativistic notations.
2.3.2. Euler-Lagrange equations

We denote as Ω the space-time domain6 on which we integrate the lagrangian
L to get the action S. The method to get the equations of motion is the same
as the one used in analytical mechanics: one should vary the action assuming
3
The case of a finite number of degrees of freedom boils down to analytical mechanics. We refer to any lecture
on analytical mechanics, as well as to the famous books of L. Landau, E. Lifchitz, Mechanics, or H. Goldstein,
Classical Mechanics.
4
This is assumed in order to satisfy locality: in order that L(x) would depend on the values of fields at point
y, distinct from x, one should be able to reconstruct the field value φ(y) from φ(x) and its derivatives at
position x, which requires generally speaking an infinite number of derivatives from Taylor exansion...
5
This is due to the Ostrogradsky instability, for theories having equations of motion with more than two time
derivatives (higher-derivative theories). A theorem of Mikhail Ostrogradsky in classical mechanics indeed
states that a non-degenerate Lagrangian dependent on time derivatives higher than the first corresponds to
a Hamiltonian unbounded from below, therefore non physical.
6
In general, Ω is the whole space-time, in which case each x component covers ] − ∞, +∞[, but it is sometime
convenient to keep Ω arbitrary in full generality, in particular in view of the Noether theorem.
29
that it should be extremal around the actual values of the fields describing the
physical system. Under the transformation
φ(x) → φ(x) + δφ(x) , (2.31)
with the constraint that fields on the boundary ∂Ω of Ω have fixed values, which
means that their variation δφ(x) vanish on ∂Ω, the action varies as
Z
δL(x) δL(x)
δS = d4 x δφi (x) + δ(∂µ φi (x)) , (2.32)
Ω δφi (x) δ∂ µ φi (x)
where we have performed a first order Taylor expansion of L, which is a func-
tional of the fields and their first derivatives. Integrating by parts, we thus
get
Z Z
δL(x) δL(x) δL(x) 3
δS = d4 x − ∂µ δφi (x) + δφi (x) d σµ (2.33)
Ω δφi (x) δ∂µ φi (x) ∂Ω δ∂µφi (x)
where d3 σµ is the infinitesimal 3-dimensional element of integration on the

boundary ∂Ω. The last term in Eq. (2.33) vanishes due to the vanishing of
the variation δφ(x) of fields on the boundary7. The action being stationary by
assumption, therefore for any field variation δφi (x), this implies from Eq. (2.33)
that
δL(x) δL(x)
− ∂µ =0 (2.34)
δφi (x) δ∂µ φi (x)
which are the Euler-Lagrange equations for the fields φi (x).
One should note that adding a 4-divergence term to the lagrangian does not
modifies the action, and therefore not the equations of motion.
2.3.3. Application: field equations of a massive scalar field

theory
Consider the lagrangian
1 1
L = ∂µΦ∂ µ Φ − m2 Φ2 . (2.35)
2 2
Its equation of motion can readily be obtained: rewriting
1 1
L = ∂µ Φg µν ∂ν Φ − m2 Φ2 , (2.36)
2 2
7
If Ω is just the entire space-time, this is consistent with the assumed vanishing values of fields at infinity.
30
2.4. Physical content
we have
δL
= −m2 Φ (2.37)
δΦ
and
δL
= ∂ µΦ . (2.38)
δ∂µ Φ
Indeed µ and ν are dummy indexes, and thus one should not forget to differen-
tiate both terms in Eq. (2.36). This finally leads to the equation of motion

µ m2 c2
∂ ∂µ + 2 Φ(x) = 0 (2.39)
~
which is our celebrated Klein-Gordon equation.

2.4.1. Probability density and current
Let us construct a conserved four-current which extends the probability den-
sity (2.16) and current density (2.17) to the relativistic case. Starting from the
Klein-Gordon equation

1 ∂2 ~ 2 m2 c2
−∇ Φ+ 2 Φ = 0, (2.40)
c2 ∂t2 ~
and multiplying by iΦ∗ , we get
1 ∗ ∂ 2Φ ∗~ 2 m2 c2 ∗
iΦ − iΦ ∇ Φ + iΦ Φ = 0 , (2.41)
c2 ∂t2 ~2
while considering the complex conjugate of the Klein-Gordon equation (2.40)
multiplied by iΦ leads to
1 ∂ 2 Φ∗ ~ 2 ∗ m2 c2 ∗
iΦ 2 − iΦ∇ Φ + 2 iΦ Φ = 0 , (2.42)
c2 ∂t ~
so that the subtraction of Eqs. (2.41, 2.42) gives
2 2 ∗

1 ∗ ∂ Φ ∂ Φ ∗~ 2 ~ 2 ∗
i Φ − Φ 2 − i Φ ∇ Φ − Φ∇ Φ = 0 , (2.43)
c2 ∂t2 ∂t
31
or equivalently

∂ ∗ ∂Φ ∂Φ∗ ~
h
∗~ ~ ∗
i
i Φ −Φ + ∇ · −i Φ ∇Φ − Φ∇Φ = 0, (2.44)
c∂t c∂t c∂t
i.e.
∂ρ ~ ~
+∇ ·j = 0, (2.45)
∂t
where
∗

i~ ∂Φ ∂Φ
ρ = 2 Φ∗ −Φ , (2.46)
c ∂t ∂t

~ − Φ∇Φ
~j = −i~ Φ∗∇Φ ~ ∗ . (2.47)
Equation (2.45) just reflects the conservation of probability if (2.46) and (2.47)
are to be interpreted as the relativistic probability density and the probability
current respectively.
Introducing the four-current

µ
j = ρc, j ,~ (2.48)
which has the covariant expression
j µ = i~ (Φ∗∂ µ Φ − Φ∂ µΦ∗) , (2.49)
the conservation of probability (2.45) which reads

∂ ~ · ~j = 0 ,
cρ + ∇ (2.50)
c∂t
can be presented in the covariant form
∂µ j µ = 0 . (2.51)
Note that the fact that ρ is the time component of a four-vector is consistent
with the fact that ρ d3 V is the probability that a particle would be in the volume
d3V . Under a boost of Lorentz factor γ, d3 V is Lorentz contracted by a factor
of 1/γ, which is compensated by the fact that ρ is dilated by a factor of γ as any
time component of a four-vector, so that the probability is indeed invariant.
32
2.4.2. Non relativistic limit

Let us check that the relativistic probability density and probability current
reduce to the non-relativistic expressions (2.5) and (2.13) respectively. Again
presenting the wave function in the form (2.26), we get

i~ mc2 ∗
ρ = −i 2Ψ Ψ + (Ψ∗∂t Ψ − Ψ∂t Ψ∗ ) , (2.52)
2mc2 ~
i~ ∗ i
ji = − Ψ ∇ Ψ − Ψ∇i Ψ∗ . (2.53)
2m
The probability current (2.53) has already the appropriate form (2.13). As far
as ρ is concerned, since the term in parenthesis in the RHS of Eq. (2.52) is
equal to −2iΨΨ∗(E − mc2 )/~, it is negligible with respect to the first term (rest
energy) in the limit c → ∞, so that ρ reduces to the non-relativistic expressions
(2.5), as expected.
2.4.3. Energy spectrum and probability density

Let us consider a plane wave solution8
Φ = N e−ip·x . (2.54)
We thus get, from Eq. (2.49),
j µ = 2pµ|N |2 . (2.55)
This looks very simple and elegant. However, substituting the plane wave solu-
tion (2.54) in the Klein-Gordon equation (2.24), we get the dispersion relation
p
2 2
p = m i.e. E = ± p~ 2 + m2 , (2.56)
so that we encounter two very serious problems:

- there are negative energy solution, and the energy spectrum is not bounded
from below, so that under an arbitrary perturbation, the system could furnish
an arbitrary amount of energy to the outside.
- the probability density being proportional to the energy, negative energy
solutions have negative probability densities!
The problem is very serious since any consistent treatment of a physical system
requires to deal with a complete set of states, as we are used to do in quantum
mechanics.
8
We use the natural units c = 1 and ~ = 1.
33
2.4.4. Charge current and reinterpretation of the negative

energy solution
It turns out that multiplying the current (2.49) by the elementary charge −e is
a way to escape the two previous problems. Let us indeed consider the current
j µ = −ie (Φ∗∂ µ Φ − Φ∂ µΦ∗) . (2.57)
The corresponding ρ = j 0 is now the charge density, and not anymore the
probability density. Leaving aside spin effects, let us use the Klein-Gordon
equation as a relativistic equation describing “electrons”. Spin effects require
the introduction of the Dirac equation, which will not be discussed in these
lectures.
Consider a plane wave (2.54). For an electron e− , of charge −e, energy E
and momentum p~, we get, using the definition (2.57) of the charge current,
j µ (e− ) = −2e|N |2 (E, ~p). (2.58)
Besides, for a positron e+ , of charge +e, of energy E and momentum p~, we get,
using the definition (2.57) of the charge current,
j µ (e+) = +2e|N |2 (E, ~p) = −2e|N |2 (−E, −~p) . (2.59)
The last algebraic manipulation may look somehow awkward and trivial. How-
ever, it just means that there is no need to introduce positron degrees of free-
dom: they are already there in the Klein-Gordon equation, since according to
the previous identity, a positron of energy E and momentum p~ is the same as an
electron of negative energy −E, of momentum −~p, i.e. propagating backward
in time.
Pictorially, we can thus draw
−−−−−−−−−−−−−−→
time
(2.60)
−→−− = −−←−
+
e e−
E>0 (−E) < 0
From the time evolution point of view of a wave function in quantum mechanics,
this relies on the fact that the time depend part of a free particle moving forward
in time, of energy E, is the same as the one of a free particle of energy −E
propagating backward in time, since
e−i(−E)(−t) = e−iEt . (2.61)
34
2.5. In-depth study: global symmetries and Noether theorem
2.5. In-depth study: global symmetries and

Noether theorem
In this section, we want to show that whenever an action is invariant under a
given global symmetry, there exists a conserved current, and a related conserved
charge. As an application, we will recover the expression of the conserved charge
current (2.57).
2.5.1. Variation of the action with respect to boundary

configurations
As a first step, let us study how the action is modified when any field config-
uration, solution of the equations of motion, faces an infinitesimal change to
another solution of the equations of motion, over the whole space-time domain
Ω, without fixing its value at the boundary δΩ. This is in contradistinction
with the situation studied previously in the section 2.3.2 when establishing the
Euler-Lagrange equation, where the boundary field configuration was fixed. We
will show that the variation of the action as a simple expression as a function
of the variation of the field configuration at the boundary.
Consider an arbitrary transformation of the coordinates and fields9
x′µ = xµ + δxµ ,
(2.62)
φ′ (x′) = φ(x) + δφ(x) .
The local variation of the field then reads
δ̄φ(x) = φ′ (x) − φ(x)

= φ(x − δx) + δφ(x − δx) − φ(x)
= δφ(x) − δxµ ∂µ φ(x) + ... (higher order terms). (2.63)
The action for fields solutions of the equation of motion, which reads, before
the transformation, Z
S= d4x L(φ(x), ∂µ φ(x), x) (2.64)
Ω
thus becomes Z
′
S = d4x L(φ′ (x), ∂µ φ′ (x), x) (2.65)
Ω′
9
Although we only consider scalar fields in the present chapter, the following discussion also applies to arbitrary
fields.
35
where Ω′ is obtained from Ω by the transformation (2.62). Besides, S ′ also reads

Z Z
′ 4 ′ ′
S = d x L(φ , ∂µ φ , x) + d3 σµ δxµ L(φ′ , ∂µ φ′ , x) , (2.66)
Ω δΩ
which gives, after a first order Taylor expansion, neglecting terms of order 2,
Z Z
′ 4 δL δL
S =S+ d x δ̄φ + ∂µ δ̄φ + d3σµ δxµ L(φ, ∂µ φ, x) + · · · .
Ω δφ δ(∂µ φ) δΩ
(2.67)
Integrating by parts, one thus gets
Z Z
′ 4 δL δL 4 δL
S = S+ d x − ∂µ δ̄φ + d x ∂µ δ̄φ (2.68)
Ω δφ δ(∂µ φ) Ω δ(∂µφ)
Z
+ d3 σµ δxµ L(φ, ∂µ φ, x) + · · · .
δΩ
The second term vanishes since fields satisfy the equations of motion. Integrat-
ing the third term, we thus get
Z
δL
S′ = S + d3 σ µ δ̄φ + δxµ L + · · · (2.69)
δΩ δ(∂µ φ)
Z
δL δL
= S+ d3 σ µ δφ − ∂ν φ δxν + δxµ L + · · ·
δΩ δ(∂µ φ) δ(∂µφ)
where we have used Eq. (2.63). Finally, we obtain

Z
3 δL δL µ ν
δS = d σµ δφ − ∂ν φ − g ν L δx + ··· , (2.70)
δΩ δ(∂µ φ) δ(∂µ φ)
after using the fact that g µν ≡ δ µν .
2.5.2. Noether theorem

In the special case where the action is invariant under the considered trans-
formations, we get the Noether theorem, which states that to any continuous
group of symmetry of the action is associated a conserved quantity.
The conserved current corresponding to this symmetry reads

µ δL δL
j = δφ − ∂ν φ − g ν L δxν
µ
(2.71)
δ(∂µ φ) δ(∂µ φ)
36
2.5. In-depth study: global symmetries and Noether theorem
since following Eq. (2.70),

Z Z
3 µ
δS = 0 = d σµ j = d4 x ∂µ j µ
δΩ Ω
whatever Ω, which indeed shows that the current j µ is conserved:
∂µ j µ = 0 .
The corresponding conserved charge reads

R
Q = d3x j 0 (x) . (2.72)
Indeed, Z Z
∂Q
= d x ∂0 j (x) = − d3x ∂i j i (x) = 0
3 0
∂t
for rapidly decreasing field at infinity. More generally, the conserved charge can
defined through the relation
Z
Q= d3σµ j µ (x) , (2.73)
N3
where N3 is a space-like 3-surface.

One should note that the above considered transformations are global (also
named rigid). We do not assume any locality of these transformations in the
proof. Imposing that the action would be invariant under local transformations
would be an additional requirement, underlying gauge field theories. It turn out
that this furnishes a dynamical principle allowing to construct various kinds of
lagrangian. This is essence of Yang-Mills construction.
2.5.3. Charge current

Let us consider again the lagrangian (2.35) of a scalar field, assuming now that
the field is complex:
L = ∂µ Φ∗ ∂ µΦ − m2 Φ∗Φ . (2.74)
It means that Φ and Φ are to be considered as independent fields. Their two

equations of motion thus read

µ m2 c2
∂ ∂µ + 2 Φ(x) = 0 (2.75)
~
37
and
2 2

m c
∂ µ ∂µ + 2 Φ∗(x) = 0 . (2.76)
~
Obviously, the lagrangian (2.74), and therefore the corresponding action, are
invariant under the so-called global gauge transforms
Φ → eieα Φ (2.77)
Φ∗ → e−ieαΦ∗ (2.78)
where e is an arbitrary parameter, which has the physical meaning of the ele-
mentary electric charge, due to its role in the conserved charge current which
we will now construct.
Let us compute the Noether current associated to this transformation. We
consider the infinitesimal versions of the transformations (2.77-2.78)
δΦ = ieδαΦ (2.79)
δΦ∗ = −ieδαΦ∗ (2.80)
with δx = 0. Therefore, according to Eq. (2.71), and using the fact that
δL
= ∂ µ Φ∗ , (2.81)
δ(∂µΦ)
δL
∗
= ∂ µΦ (2.82)
δ(∂µ Φ )
we get
j µ = ∂ µ Φ∗(ieδαΦ) − ∂ µ Φ(ieδαΦ∗) (2.83)
valid for any δα, which can thus be factorized out, so that the electromagnetic
four-current
j µ = −ie (Φ∗ ∂ µΦ − Φ ∂ µΦ∗) (2.84)
is conserved. We have therefore recovered, now from a global symmetry prin-

ciple, the current (2.57) which was obtained in a heuristic way.
38
Part II.
Introduction to Quantum
Electrodynamics
39
3. From nonrelativistic
perturbation theory to Feynman
diagrams
3.1. Time dependent perturbation theory
General framework
We consider a particle which is described by the hamiltonian
H = H0 + V (t) , (3.1)
where H0 is the free hamiltonian which has known spectrum and eigenstates,
and V (t) is an interaction potential, which generally speaking depends on time.
(0)
The eigenstates of the free hamiltonian H0 are denoted as |Ψn i. They thus
satisfy
H0 |Ψ(0) (0)
n i = En |Ψn i (3.2)
with the orthonormalization properties
hΨ(0) (0)
n |Ψm i = δnm . (3.3)
The solutions of the free Schrödinger equation

∂ (0)
H0 |Ψ(0)
n (t)i = i~ |Ψ (t)i (3.4)
∂t n
are thus given by
En t
|Ψ(0)
n (t)i = e
−i ~ |Ψ(0)
n i. (3.5)
We want to solve the full Schrödinger’s equation governed by the hamiltonian

H
∂
H|Ψ(t)i = i~ |Ψ(t)i . (3.6)
∂t
41
3. From nonrelativistic perturbation theory to Feynman diagrams
For that, we aim at finding its solution through an expansion over the basis
(0)
|Ψn (t)i of solutions of the free Schrödinger’s equation (3.4) governed by the
hamiltonian H0 .
We thus generically write
X X En t
|Ψ(t)i = an (t) |Ψ(0)
n (t)i = an (t) e−i ~ |Ψ(0)
n i. (3.7)
n n
In terms of wave functions, this reads

X En t
X En t
Ψ(~x, t) = h~x|Ψ(t)i = an (t) e−i ~ h~x|Ψ(0)
n i = an (t) e−i ~ Ψ(0) (~x, t) . (3.8)
n n
The problem thus reduces to the determination of the coefficients an (t).
Inserting the expansion (3.7) inside the Schrödinger’s equation (3.6), we ob-
tain
∂ X dan (t) En t
−i ~ (0)
X En t
i~ |Ψ(t)i = i~ e |Ψn i + an (t)Ene−i ~ |Ψ(0)
n i (3.9)
∂t dt
Xn n
X
−i E~nt (0) En t
= an (t)e V (t)|Ψn i + an (t)e−i ~ H0 |Ψ(0)
n i (3.10)
n n
and thus
X dan (t) En t
X En t
i~ e−i ~ |Ψ(0)
n i = an (t)e−i ~ V (t)|Ψ(0)
n i, (3.11)
n
dt n
which is up to now an exact equation.

Before proceeding further, one should note that what we have done is noth-
ing more than the method of variation of constant for a 1st order differential
equation: indeed the Schrödinger’s equation is just a (complicate) 1st order
differential equation...!
(0)
Contracting Eq. (3.11) on the left with hΨf | leads to
X dan (t) En t (0) −i X En t (0)
e−i ~ hΨf |Ψ(0)
n i = an (t)e−i ~ hΨf |V (t)|Ψ(0)
n i , (3.12)
n
dt ~ n
and thus to the set of coupled linear differential equations
daf (t) −i X (0)

(E −En )t
i f~
= an (t) hΨf |V (t)|Ψ(0)
n i e . (3.13)
dt ~ n
42
To proceed, we now make an assumption: much before the potential V starts

(0)
to act, the system is in the state |Ψi i, say at time −T /2, with T very large.
Thus,
an (−T /2) = δni . (3.14)
The set (3.13) can be equivalently represented in integral form, namely

Z
i t ′X ′ (0)
(E −En )t′
(0) i f ~
af (t) = af (−T /2) − dt an (t ) hΨf |V (t)|Ψn i e . (3.15)
~ − T2 n
In general, this cannot be solved exactly.
Perturbation theory
We now assume that V is small, so that a perturbative treatment can be per-
formed. This is nothing more than a sophisticated series expansion in powers
of V . We thus write
(0) (1) (2)
af (t) = af + af + af + ···
order V 0 order V 1 order V 2 (3.16)
Let us examine the first terms in this expansion.

The generic term is obtained through the expansion
(0) (1) (2)
af (t) = af + af + af + · · · (3.17)
Z
i t ′ X (0) ′
(E −En )t′
(0) i f ~
= af (−T /2) − dt hΨf |V (t )|Ψn i e (3.18)
~ − T2 n
h i
(0) ′ (1) ′ (2) ′
× an (t ) + an (t ) + an (t ) + · · · (3.19)
The order 0 term is trivial: it is solved by setting V = 0, i.e.

(0)
af (t) = af (−T /2) = δf i . (3.20)
The order 1 term reads:

Z t
(1) i ′
X (0) ′
(Ef −En )t′
af (t) = − dt hΨf |V (t )|Ψ(0)
n ie
i ~ δni (3.21)
~ − T2 n
Z t
i ′ (0) ′ (0)
(Ef −Ei )t′
= − dt hΨf |V (t )|Ψi i ei ~ (3.22)
~ − T2
43
The order 2 term reads:

2 Z t
(2) i ′
X (0)
′
(E −En )t′
(0) i f ~
af (t) = − dt hΨf |V (t )|Ψn i e
~ −2T
n
Z t′ ′′
′′ (0) ′′ (0) i (En −E i )t
× dt hΨn |V (t )|Ψi i e ~ (3.23)
− T2
etc.
Transition amplitude
We are mostly interested in transitions between an initial state (t = −T /2) and
a final state (t = +T /2), in which the interaction can be neglected both in the
initial and in the final states. We therefore define the transition amplitude
Z T
i 2
′
X
′ (0)
(Ef −En )t′
Tf i = af (T /2) − af (−T /2) = − dt an (t ) hΨf |V (t)|Ψ(0)
n ie
i ~ ,
~ − T2 n
(3.24)
where we have subtracted the contribution af (−T /2), since we are not inter-
ested in the situation in which nothing occurred.
The potential V (t) is assumed to be a local space-time operator, i.e.
h~x|V (t)|~x ′ i = V (x)δ (4)(x − x′ ) . (3.25)
Inserting twice the identity operator

Z
1 = d3 ~x |~xih~x| (3.26)
inside (3.24), we get
Tf i (3.27)
Z T Z Z
i 2 X (0)
(Ef −En )t
=− dt d3 ~x d3 ~x ′ an (t) hΨf |~xih~x|V (t)|~x ′ ih~x ′ |Ψ(0)
n i an (t) e
i ~
~ − T2 n
Z T Z
i 2
∗(0)
X
=− dt d3 ~x Ψf (x) V (x) Ψ(0)
n (x)an (t) ,
~ − T2 n
where we have used the fact, see Eq.(3.5), that

En t
Ψ(0) (0)
x) = Ψ(0)
n (x) ≡ Ψn (t, ~ x) e−i
n (0, ~
~ . (3.28)
44
In the T → ∞ limit, we thus have the covariant expression

Z
i 4 ∗(0)
X
Tf i = − d x Ψf (x) V (x) Ψ(0)
n (x) an (t) . (3.29)
~ n
First order transition amplitude for scattering on a static potential

Let us focus now on the first order expansion in V . We thus have
Z T →+∞
(1) i 2
(0) (0)
Ef −Ei
Tf i = − dt hΨf |V (t)|Ψi i ei ~ t
~ − T2 →−∞
Z
i ∗(0) (0)
= − d4x Ψf (x) V (x)Ψi (x) . (3.30)
~
We now consider the special situation in which V is time independent:
V (t, ~x) = V (~x) . (3.31)
Thus,
Z +∞
(1) i Ef −Ei
Tf i = − Vf i dt ei ~ t
= −2πi Vf i δ(Ef − Ei) . (3.32)
~ −∞
where
Z
(0) (0) ∗(0) (0)
Vf i = hΨf |V |Ψi i = d3~x Ψf (~x) V (~x) Ψi (~x) . (3.33)
One should note that assuming V to be time-independent is equivalent to have

no energy transfer from the potential to the particle, since integration over time
leads to a δ(Ef − Ei) .
Thus, we recover the fact for a static potential, on which the scattering takes
an infinite amount of time,
∆t = ∞ ⇔ ∆E = 0 (3.34)
in accordance with Heisenberg’s uncertainty ∆t∆E & ~/2 .
Fermi’s Golden rule

We are interested in the probability that the particle is scattered on V from an
initial state i to a final state f . Quantum mechanics tells us that one should
square Tf i, but we immediately face a serious technical problem, since this
implies to compute the square [δ(Ef − Ei )]2 which looks like meaningless.
45
The solution is to compute the transition probability per unit of time

|Tf i|2
Wf i = lim . (3.35)
T →∞ T
The trick is to write [δ(Ef − Ei )]2 in a non symmetric way:

Z T
(1) 2 2π (1) 2 2 Ef −Ei
|Tf i | = |Vf i | δ(Ef − Ei) dt ei ~ t
~ − T2
Z T
2π (1) 2 2
= |Vf i | δ(Ef − Ei) dt
~ − T2
2π (1) 2
= |V | δ(Ef − Ei) T (3.36)
~ fi
and thus
2π (1) 2
Wf i = |V | δ(Ef − Ei ) . (3.37)
~ fi
Note: one can easily check that W has the correct dimension. Since [~] = [E] T,
[E]2 [E]
with [Vf2i δ(Ef − Ei )] = = [E] we thus have[W ] = = T −1.
[E] [E] T
In practice, one usually deals with final states in a given set. We therefore
introduce the density ρ(Ef ) of these states, so that ρ(Ef ) dEf is the number of
states in the energy range Ef to Ef + dEf . Thus, the transition probability per
unit time Wf i reads
Z
2π (1)
Wf i = dEf ρ(Ef ) |Vf i |2 δ(Ef − Ei ) , (3.38)
~
so that
2π (1) 2
Wf i = |V | ρ(Ei) , (3.39)
~ fi
known under the name of Fermi’s Golden rule.
Order 2
(2)
From the expression (3.23) obtained for af (t), one gets
2 Z T
(2) i 2 X (0) E −En
i f ~ t′
Tf i = − dt′ hΨf |V (t′ )|Ψ(0) n i e
~ − T2 n
Z t′
(0) i En ~−Ei t′′
× dt′′ hΨ(0)
n |V (t′′
)|Ψ i ie . (3.40)
− T2
46
Again working under the assumption of a time-independent potential, we have

2 X Z T Z t′
(2) i 2
′ i
Ef −En ′
t
En −Ei ′′
Tf i = − Vf n Vni dt e ~ dt′′ ei ~ t
. (3.41)
~ n − T2 − T2
We should now take the limit T → ∞. Meanwhile, the interaction should cease
at t = −∞ and t = +∞. For that, we introduce an adiabatic parameter ǫ > 0
to make the t′′ integration meaningful. Physically, it is introduced in order to
circumvent the fact that choosing a time-independent potential is at odd with
the fact that it is supposed to vanish at t = −∞ and t = +∞.
One now has
Z t′
En −Ei −iε ′′
dt′′ ei ~ t
− T2
1 h En−Ei −iε ′ E −E −iε
i
i t i n ~ i (− T2 )
= i~ e ~ −e
Ei − En + iε
1 En −Ei −iε ′
−−−−−→ i~ ei ~ t (3.42)
T →+∞ Ei − En + iε
where we have used the fact that second term in the second line vanishes in the
limit T → +∞, thanks to the damping factor exp(−εT /(2~)).
Thus,
2 X Z +∞
(2) i Vf n Vni Ef −En +En −Ei ′
Tf i = − i~ dt′ ei ~ t
~ n
Ei − En + iε −∞
X 1
= −2πi Vf n Vni δ(Ei − Ef ) , (3.43)
n
E i − E n + iε
which exhibit the structure of the result: the transition amplitude is the product
of two matrix elements of the potential, between which an energy denominator
is to be inserted, describing the propagation of the particle from state i to
state f . The concept of propagation can be made more precise based on the
formalism of Green’s functions, see any advanced quantum mechanics lectures,
e.g. Messiah Chap. XIX.
3.1.1. Higher orders

The same line of thought can be pursued at order 3. In this case, 3 matrix
(3)
elements of the potential and two energy denominators will be involved in Tf i .
Finally, the general structure is the following
47
(1) (2) (3)

Tf i = Tf i + Tf i + Tf i + · · · (3.44)
"
X 1
= −2πi Vf i + Vf n Vni
n
E i − E n + iε
#
X 1 1
+ Vf n Vnm Vmi + · · · δ(Ei − Ef ) ,
n,m
Em − En + iε Ei − Em + iε
This is illustrated at order 1 in Fig. 3.1.
•V
fi
Figure 3.1.: Perturbative expansion of Tf i at 1st order, drawn in space-time.
Fig. 3.2 illustrates the scattering at order 2. A possibility of producing a

pair of virtual particles appears. Indeed one should pay attention to time
ordering: reading Eq. (3.43) in normal order, i.e. from left to right corresponds
to going backward with respect to the direction of the arrows. The physical
interpretation of the configurations encountered at each step depends on the
ordering of time:
⋄ in the first diagram, following this backward flow of arrows, one first faces
the last scattering (at time t2 ) and then the first scattering (at time t1 ).
Therefore, this first diagram involves a single particle at every stage.
⋄ in the second diagram, following this backward flow of arrows, one first
faces the first scattering (at time t1 ) and then the second scattering (at
time t2 ).
As we have seen in Chap. 2, a particle propagating back in time should be
interpreted as an antiparticle propagating forward in time. At time t1 , a
particle-antiparticle pair is produced, so that between the two scatterings
48
t
f f
t2 Vf n • n Vf n n •
t1 • Vni • Vni
i i
Figure 3.2.: Perturbative expansion of Tf i at 2nd order, drawn in space-time,

with two diagrams corresponding to two different time orderings of
the double scattering of the electron on the external potential. In
the second diagram, a virtual pair is produced in the first scattering
(at time t1 ), which is annihilated in the second scattering (at time
t2 ).
at t1 and then t2 , the state is made of two particles and one antiparticle.
At time t2, the virtual pair annihilates, and the state is made of a single
particle.
49
4. Electrodynamics
We use the Heaviside-Lorentz’s units: Coulomb force is given by
QQ′
F = (4.1)
4πr2
which means that ε0 = 1 and µ0 = 1 . Furthermore, we use a system of units in
which c = 1 . We refer to Jackson’s appendix1 for a detailed discussion on the
various systems of units.
4.1. Maxwell equations

4.1.1. Local form
We recall that in local form, the 4 Maxwell equations read
~
divE = ρ, (4.2)
−→ ~ ∂E ~
rotB − = ~j , (4.3)
∂t
~
divB = 0, (4.4)
−→ ~ ∂B ~
rotE + = 0. (4.5)
∂t
The local conservation of charges states that
∂ρ
+ div~j = 0 . (4.6)
∂t
4.1.2. Integral form

Their corresponding integral form are:
⋄ Gauss law: for any closed surface S enclosing a volume V , one has
ZZ ZZZ
E~ · d2 S
~= ρ d3 x . (4.7)
S V
1
J. D. Jackson, Classical Electrodynamics.
51
4. Electrodynamics
⋄ Generalized Ampère’s law: for any closed curve C bordering a surface S,

I ZZ !
~
~ · d~ℓ =
B ~j + ∂ E · d2S~, (4.8)
C S ∂t
~
where ∂∂tE is the displacement current, which was absent in the original
Ampère’s law.
⋄ Absence of magnetic monopole:
for any closed surface S,
ZZ
~ · d2 S
B ~ = 0, (4.9)
C
which states that there is no magnetic monopole, in contradistinction with

Gauss’s law which states that electric charges exists.
⋄ Faraday’s induction law:
for any closed curve C, bordering a surface S,
ZZ ZZ ~
~ · d~ℓ = ∂B 2 ~
E ·d S. (4.10)
C S ∂t
4.2. Covariant formulation of electrodynamics

As usual, we denote a space-time point in Minkowski space as
xµ = (t, ~x) (4.11)
and the 4-electric current as

j µ = (ρ, ~j) . (4.12)
4.2.1. Field strength tensor and its dual

In contra-variant form, the field strength tensor is an antisymmetric tensor,
defined as
 
0 −E 2 −E 2 −E 3
µν
 E1 0 −B 3 B2 
F = 2  (4.13)
E B3 0 −B 1 
E 3 −B 2 B1 0
52
i.e.
F i0 = −F 0i = −Fi0 = F0i = E i = −Ei , (4.14)

F ij = Fij = εijk Bk = −εijk B k . (4.15)
Conversely, the magnetic field can be expressed from
εijk Fij = εijk εijk′ Bk′ = 2Bk (4.16)
since εijk εijk′ = 2δkk′ . Thus,

1
Bk = εijk Fij = −B k . (4.17)
2
Another useful tensor is obtained from F µν , its dual. To define it, one should
introduce the 4-dimensional Levi-Civita tensor ε (see Appendix B for details).
Its is defined as a fully antisymmetric tensor such that its contra-variant com-
ponents are given by2

 +1 if {µνρσ} is an even permutation of {0,1,2,3}
µνρσ
ε = −1 if {µνρσ} is an odd permutation of {0,1,2,3} (4.18)

0 otherwise.
Since det g = −1, we thus have
εµνρσ = −εµνρσ . (4.19)
The dual tensor is then defined as
1
F̃ µν = εµνρσ Fρσ = −F̃ νµ (4.20)
2
since εµνρσ is completely antisymmetric.
Conversely,
1
εµνρσ F̃ ρσ = εµνρσ ερσλτ Fλτ
2
1
= ερσµν ερσλτ Fλτ
2
1
= (−2)(gµλgντ − gµτ gνλ )Fλτ = −2Fµν (4.21)
2
2
Note that this definition is not universal. Some authors use this definition for the covariant components,
which leads to an opposite tensor with respect to the present definition.
53
4. Electrodynamics
and thus
1
Fµν = − εµνρσ F̃ ρσ . (4.22)
2
In conclusion, we have
1 µνρσ
F̃ µν = ε Fρσ (4.23)
2
1
Fµν = − εµνρσ F̃ ρσ . (4.24)
2
The components of F̃ µν can be obtained easily:

1 1 1
F̃ 0i = ε0iρσ Fρσ = ε0ijk Fjk = εijk Fjk = Bi = −B i , (4.25)
2 2 2
and
1 1 1
F̃ ij = εijρσ Fρσ = εij0k F0k + εijk0Fk0 = ε0ijk F0k = εijk F0k = εijk E k .(4.26)
2 2 2
Thus,
 
0 −B 1 −B 2 −B 3
 B1 0 E 3 −E 2 
F̃ µν = 
 B 2 −E 3
 (4.27)
0 E1 
B3 E 2 −E 1 0
i.e.
F̃ i0 = −F̃ 0i = −F̃i0 = F̃0i = B i = −Bi , (4.28)

F̃ ij = F̃ij = −εijk Ek = εijk E k . (4.29)
Besides, the two tensors are related through the following transformation:
~ → −E
B ~
µν
F −−−−−→ F̃ µν . (4.30)
~ →B
E ~
4.2.2. Lorentz transformations

As for any contra-variant 2-tensor, F µν transforms under a Lorentz transfor-
mation Λ as
= Λµµ′ Λνν ′ F µ ν .
′ ′ ′
µν
F (4.31)
54
Orthogonal transformations
Le us first consider orthogonal transformations O(3). They correspond to par-
ticular transformations Λ of the form
Λ00 = 1 Λ0i = Λi 0 = 0 Λi j = Oi j (4.32)
i.e.
 
1 0 0 0
0 
Λ= . (4.33)
0
0
O 
The 3 × 3 matrix O satisfies the constraints
Ot O = OOt = Id (4.34)
by definition of O(3), which is the set of matrices of real orthogonal matrices,

which equivalently leave the scalar product ~x · ~y invariant.
Taking the determinant of Eq. (4.34), one gets
(det O)2 = 1 (4.35)
so that:
⋄ either det O = +1: SO(3) which are rotations
⋄ either det O = −1: O = R · P where R ∈ SO(3) and P is the parity:

P = −Id3×3.
From Eq. (4.34), we get the following relations among components of O :
′ ′ ′ ′
(Ot )i j ′ Ojj = Oji Ojj = δij and (O)i j ′ (Ot )j j = Oi j ′ Ojj ′ = δij . (4.36)
Let us now consider the way various components of F transform under the
Lorentz transform Λ.
⋄ Electric field:
′
i0
F = Λi j Λ00 F j0 , (4.37)
so that
′
E i = Oi j E j (4.38)
55
4. Electrodynamics
~ transforms as a
as expected for a vector: this equation tells us that E
vector under a rotation O = R ∈ SO(3), and gets reversed, just as ~x,
under parity.
Therefore
~ is a vector (also named polar vector).

E
⋄ Magnetic field:
F ′ ij = Λi i′ Λj j ′ F i j ,
′ ′
(4.39)
with
′
F ′ ij = −εi′ j ′ k′ B k . (4.40)
We thus have
1 1
B ′ k = − εijk F ′ ij = εijk Oi i′ Ojj ′ εi′ j ′ k′ B k .
′
(4.41)
2 2
Using the fact that
εijk = εijnδnk (4.42)
and
Onp Okp = δnk , (4.43)
see Eq. (4.36), we thus get
εijk Oi i′ Ojj ′ = εijn Oi i′ Ojj ′ Onp Okp = εi′ j ′ p Okp det O , (4.44)
since
εijnOi i′ Ojj ′ Onp = εi′j ′ p det O , (4.45)
see Appendix B. We finally get
1 1 ′ ′
B ′ k = − εijk F ′ ij = εi′j ′ p εi′j ′ k′ Okp B k det O = det O Okk′ B k , (4.46)
2 2
since εi′ j ′ p εi′ j ′ k′ = 2δpk′ .
~
Due to the presence of the prefactor det O, this equation tells us that B
transforms as a vector under a rotation O = R ∈ SO(3), and remains
invariant under parity. Therefore,
~ is a pseudo-vector (also named axial vector).

B
56
Pure boosts
One can show, see tutorial, that under an arbitrary boost along the direction
~n (~n2 = 1), i.e. with a velocity ~v = β~n,
h i
~ ′ ~ ~ ~
E = (E · ~n)~n + γ E − (E · ~n)~n + γ ~v ∧ B ~, (4.47)
h i
~ ′ = (B
B ~ · ~n)~n + γ B~ − (B
~ · ~n)~n − γ ~v ∧ E
~. (4.48)
In the non-relativistic limit, the transformations (4.47) and (4.48) simplifies

into
~′ = E
E ~ + ~v ∧ B
~, (4.49)
~′ = B
B ~ − ~v ∧ E
~, (4.50)
which are particularly important for induction.
4.2.3. Relativistic invariants

From the two tensors F and F̃ , on can easily build two Lorentz invariants.
⋄ First, consider the contraction of F with itself:
Fµν F µν = F0iF 0i + Fi0F i0 + Fij F ij

= 2F0iF 0i + εijk Bk εijk′ Bk′
= −2E iE i + 2Bk Bk
= −2E~ 2 + 2B~ 2 = −2(E ~2−B~ 2) . (4.51)
Similarly,
1 ′ ′
F̃µν F̃ µν = εµνρσ F ρσ εµνρ σ Fρ′ σ′
4 ′ ′
1 σ ′ ρ′
= (−2) gρ gσ − gρ gσ F ρσ Fρ′ σ′ = −F ρσ Fρσ , (4.52)
ρ σ
4
therefore identical (up to the sign) with the contraction of F with itself,
see the expression (4.51), a fact which is obvious by just applying the
symmetry (4.30) to the expression (4.51).
57
4. Electrodynamics
⋄ Second, consider the contraction of F with F̃ :

1
Fµν F̃ µν = Fµν εµνρσ Fρσ
2
1 1 1 1
= Fij εijk0Fk0 + Fij εij0k F0k + F0iε0ijk Fjk + Fi0εi0jk Fjk
2 2 2 2
0ijk 0ijk
= −Fij ε Fk0 + F0iε Fjk
= −Fij εijk Fk0 + F0iεijk Fjk
= 2F0iεijk Fjk = −4E iB i = −4E ~ ·B
~. (4.53)
In conclusion, we have built two relativistic invariants from the two tensors F
and F̃ :
~2−B
Fµν F µν = −F̃µν F̃ µν = −2(E ~ 2) , (4.54)
~ ·B
Fµν F̃ µν = −4E ~. (4.55)
Note that under the transformation (E, ~ B)~ → (B,~ −E),

~ the first invariant (4.54)
is unchanged, while the second one (4.55) gets a minus sign, in accordance with
(4.30). Indeed, the invariant (4.54) is a Lorentz scalar, while the invariant (4.55)
is a Lorentz pseudo-scalar.
4.3. Covariant form of Maxwell’s equations

4.3.1. Maxwell’s equations
We will now show that the four Maxwell’s equations (4.2) can be summarized
in the following very elegant set of two equations, which are explicitly covariant
∂µ F µν = j ν , (4.56)
∂µ F̃ µν = 0 . (4.57)
Proof:
⋄ Gauss’s law:
→ ~
−
∂i F i0 = j 0 ⇔ ∇ · E =ρ (4.58)
58
⋄ Generalization of Maxwell-Ampère’s law:
∂0F 0k + ∂iF ik = j k (4.59)
with
F ik = −εikn B n and F 0k = −E k (4.60)
so that
~ →i ~ n
− ~ → ~ k
−
k
−∂t E − εikn ∇ B = −∂t E + ∇ ∧ B = j k (4.61)
i.e.
−→ ~ ∂E ~
rotB − = ~j . (4.62)
∂t
⋄ Absence of magnetic monopole:
∂i F̃ i0 = 0 ⇔ ∂i B i = 0 ⇔ ~ = 0.
divB (4.63)
⋄ Maxwell-Faraday’s equation:
∂0F̃ 0k + ∂iF̃ ik = 0 . (4.64)
with
F̃ ik = εiknE n and F̃ 0k = −B k (4.65)
so that
−→~ B ~
rotE + = 0. (4.66)
∂t
Compatibility condition:
The field strength tensor F µν is antisymmetric, therefore
∂ν ∂µ F µν = 0 = ∂ν j ν , (4.67)
which means that the current should be conserved:
∂ν j ν = 0 . (4.68)
59
4. Electrodynamics
4.3.2. Four-potential
From F µν to Aµ
Several remarks are in order:
~ and B,
⋄ Maxwell’s equations, expressed in terms of E ~ are not covariant.
⋄ They only involve first order time derivative of the dynamical variables
~ and B.
E ~ Thus, their conjugated momenta are not independent of the
dynamical variable3.
One should rather look for second order equations.
Let us introduce the four-potential Aµ such that
F µν = ∂ µAν − ∂ ν Aµ . (4.69)
Theorem (Poincaré):
The equation ∂µ F̃ µν = 0 is a necessary condition for Aµ to exist. This is

a sufficient condition if the space is contractible, which means that it can be
smoothly contracted to a point.
Eq. (4.69) implies that
→ 0 ∂A
− ~
F i0 i 0
= ∂ A −∂ A 0 i
⇔ ~
E = − ∇A − (4.70)
∂t
and
F ij = ∂ iAj − ∂ j Ai (4.71)
i.e.
1 1 → ~ k
−
ij i j j i i j
− εijk F = − εijk ∂ A − ∂ A = −εijk ∂ A = ∇ ∧ A (4.72)
2 2
and thus
→ 0 ∂A
− ~
~
E = − ∇A − (4.73)
∂t
→
−
~ = ∇ ∧A~.
B (4.74)
3
Consider for example the trivial case L = φ̇φ, an action in which φ̇ does not appear quadratically.
60
Consistency check:
One should have ∂µ F̃ µν = 0 which is indeed satisfied since

1
∂µ F̃ µν = εµνρσ ∂µ (∂ρ Aσ − ∂σ Aρ ) = εµνρσ ∂µ ∂ρ Aσ = 0 , (4.75)
2
since εµνρσ is symmetric under µ ←→ ρ exchange, while ∂µ ∂ρ is symmetric.
Gauge invariance
The relation (4.69) does not provide a unique 4-potential Aµ .
Indeed, the gauge transformation
Aµ (x) → Aµ(x) + ∂ µ φ , (4.76)
where φ is an arbitrary function, leaves F µν invariant.
Assuming the space-time domain to be contractible, Maxwell’s equations are
equivalent to
∂µ F µν = Aν − ∂ ν (∂µ Aµ ) = j µ . (4.77)
The equation
Aν − ∂ ν (∂µ Aµ ) = j µ (4.78)
is gauge invariant, since under the gauge transformation (4.76), we have
Aν − ∂ ν (∂µ Aµ ) → Aν + ∂ ν φ − ∂ ν (∂µ Aµ ) − ∂ ν φ = Aν − ∂ ν (∂µ Aµ ) .
(4.79)
Let us write the various components of Maxwell’s equations (4.78) explicitly:
⋄ Time component:

0 ∂ ∂A0 ~ = 0,
A − + divA (4.80)
∂t ∂t
i.e.
∂ ~ = ρ.
−∆A0 − divA (4.81)
∂t
⋄ Space components:
0
−
→
~+∇ ∂A ~ = ~j .
A + divA (4.82)
∂t
61
4. Electrodynamics
Usual gauges
There is an infinite set of possible gauge choices. The most popular one are:
⋄ Landau gauge
It relies on the covariant gauge-fixing condition
∂µ Aµ = 0 (4.83)
which thus implies that
Aν = j ν . (4.84)
⋄ Coulomb gauge
It relies on the non-covariant gauge-fixing condition
~=0
divA (4.85)
which implies that the scalar potential A0 satisfies
−∆A0 = ρ . (4.86)
The solution of this second-order differential equation is given by
Z
1 ρ(t, ~y)
A0 (t, ~x) = d3 y (4.87)
4π |~y − ~x|
since
1 ′
−∆ = 4πδ(~
r − ~
r ). (4.88)
|~r − ~r ′ |
In this gauge, A0 is therefore an instantaneous potential. The vector
~ is then, from Eq. (4.82), solution of the equation
potential A
Z 3 ′ ′
→
−
~ = ~j − ∇ d x ∂ρ(t, ~
x ) 1
A . (4.89)
4π ∂t |~x − ~x ′ |
Note the consistency with the fact that the current should be conserved:
Z 3 ′ ′
~ = div~j − ∆ d x ∂ρ(t, ~
x ) 1
div A (4.90)
4π ∂t |~x − ~x ′ |
Z ′
′ ∂ρ(t, ~x )
= div~j + d3x′ δ(~x − ~x ) (4.91)
∂t
∂ρ(t, ~x)
= div~j + = 0, (4.92)
∂t
which is therefore consistent with our gauge choice divA ~ = 0.
62
4.4. Lagrangian for photons
⋄ Temporal gauge
In this gauge, the scalar potential is chosen to vanish
A0 = 0 . (4.93)
⋄ Axial gauge
In this gauge, specified by a given space-like direction, e.g. z,
A3 = 0 . (4.94)
⋄ Light-cone gauge
This gauge, of particular use in high-energy particle physics, is specified
by choosing a light-cone vector n with n2 = 0, such that
A· n = 0. (4.95)
⋄ General axial gauge: it includes the three previous cases, by introducing

a vector n, imposing
A· n = 0. (4.96)
Clearly, n2 > 0 is the temporal gauge, n2 < 0 is the pure axial gauge, and
n2 = 0 is the light-cone gauge.

We are looking for a lagrangian describing the photon degrees of freedom. It
should be local, quadratic in the field Aµ and its derivatives, Lorentz invariant,
and its equations of motion should be (4.78). We thus generically write
L(x) = a Aµ Aµ + b (∂µAν )(∂ν Aµ ) + c (∂µ Aν )(∂ µAν ) + d (∂µ Aµ )2 + e Aµ j µ .(4.97)
The Euler-Lagrange equation

∂L ∂L
− ∂ σ =0 (4.98)
∂Aρ ∂(∂σ Aρ )
then leads to

2aAρ + ejρ = ∂σ 2b ∂ρAσ + 2c ∂ σ Aρ + 2d δρσ ∂µ Aµ (4.99)
63
4. Electrodynamics
Let us study the gauge invariance of the equation of motion for j = 0 : under
the gauge transformation (4.76), we get
a ∂ρ φ = b ∂ρ∂σ ∂ σ φ + c ∂ρ ∂σ ∂ σ φ + d ∂ρ ∂σ ∂ σ φ = (b + c + d)∂ρφ , (4.100)
therefore one should have a = 0 and b + c + d = 0 .

Besides, rewriting the term (∂µ Aµ )2 in the Lagrangian density as
(∂µ Aµ )2 = (∂µAν )(∂ν Aµ) + ∂µ [Aν (g µν ∂ρ Aρ − ∂ ν Aµ )] , (4.101)
and using the fact that the second term in the r.h.s. is a total derivative, which
thus does not contribute to the action, it can be simply omitted. From the fact
that b + d = −c, we get
L = −c (∂µ Aν ∂ν Aµ − ∂µ Aν ∂ µ Aν ) + e Aµ j µ . (4.102)
Using the fact that

1
∂µ Aν ∂ν Aµ − ∂µ Aν ∂ µAν = − F µν Fµν , (4.103)
2
the Lagrangian density thus reads
c
L = F µν Fµν + e Aµ j µ . (4.104)
2
It remains to adjust the values of c and e properly.
Let us come back to the equation of motion (4.98). Using the fact that
∂
ρ
(Fµν F µν ) = 4F σρ , (4.105)
∂(∂σ A )
we get
c
4 ∂σ F σρ = e jρ (4.106)
2
so that in order for this equation to be equivalent to the Maxwell equation (4.56),
one should have 2c = e.
Actually, the overall normalization, as well as the relative value of c and e
can be obtained as follows:
~ is a dynamical variable. Thus, the ’kinetic’ term should be
⋄ A
!2
1 ∂A ~
(4.107)
2 ∂t
64
in L. Now, by inspection
c
Fµν F µν (4.108)
2
provides
c 2
× 2 × F 0i F0i = c ∂ 0 Ai∂0Ai = −c ∂0 Ai (4.109)
2
so that c = − 21 .
⋄ In the Lagrangian density L = T − U where T is the kinetic energy and

U is the potential energy, we know that U should contain the potential
energy ρA0 in exactly this form. Besides, j µ Aµ provides exactly ρA0 , so
that e = −1.
In conclusion,
L(x) = − 41 Fµν F µν − j µ Aµ , (4.110)
and the corresponding action reads

Z
R 1 ~ 2 ~ 2 ~ .
I= L(x) d4x = 4
dx E − B − ρA + ~j · A
0
(4.111)
2
One should note that the Lagrangian density (4.110) is not gauge invariant if
j(x) is an external current. Indeed under a gauge transformation,
Aµ → Aµ + ∂ µ φ
L(x) −−−−−−−→ L(x) − jµ ∂ µ φ . (4.112)
Still, since
jµ ∂ µφ = ∂ µ (φjµ ) − φ ∂ µ jµ (4.113)
in which, in the r.h.s, the first term is a total derivative and the second one van-
ishes because jµ is a conserved current. Therefore, both terms can be omitted
from the action I.
65
4. Electrodynamics
4.5. Coupling between matter and

electromagnetic field
4.5.1. Matter part
As a reminder, consider two real scalar fields φ1 and φ2 , of identical masses.
The Lagrangian densities for these two fields are
1 1 2 2
L1 = (∂µ φ1 ) (∂ µφ1 ) − m φ1 , (4.114)
2 2
1 1 2 2
L2 = (∂µ φ2 ) (∂ µφ2 ) − m φ2 . (4.115)
2 2
Introducing the fields
1
Φ = √ [Φ1 + i Φ2 ] , (4.116)
2
1
Φ∗ = √ [Φ1 − i Φ2] (4.117)
2
which are treated as independent fields, we have
1 2
|Φ|2 = Φ Φ∗ = Φ1 + Φ22 (4.118)
2
and
|∂µ Φ|2 = (∂µ Φ) (∂ µ Φ)∗ (4.119)

1
= (∂µ Φ1 + i ∂µ Φ2) (∂ µ Φ1 − i ∂ µΦ2 ) (4.120)
2
1 1
= (∂µ Φ1) (∂ µ Φ1) + (∂µ Φ2 ) (∂ µΦ2) . (4.121)
2 2
Thus, L = L1 + L2 can be rewritten as
L = (∂µ Φ) (∂ µ Φ)∗ − m2 Φ Φ∗. (4.122)
The Euler-Lagrange equations, which read

∂L ∂L
− ∂σ = 0 ⇒ Φ∗ + m2 Φ∗ = 0 (4.123)
∂Φ ∂(∂σ Φ)
∂L ∂L
∗
− ∂ σ ∗
= 0 ⇒ Φ + m2 Φ = 0 (4.124)
∂Φ ∂(∂σ Φ )
66
4.5. Coupling between matter and electromagnetic field
As we have seen in Chap. 2, under global U (1) transformations
Φ → eieα Φ i.e. δΦ = ieδαΦ , (4.125)

Φ∗ → e−ieα Φ∗ i.e. δΦ∗ = −ieδαΦ∗ , (4.126)
the Lagrangian L is invariant, which implies that there is a conserved Noether

current, namely

1 ∂L ∂L
jµ = δΦ + ∗
δΦ∗
δα ∂(∂µΦ) ∂(∂µΦ )
= ie [(∂ Φ )Φ − (∂ Φ)Φ∗]
µ ∗ µ
= −ie [Φ∗ (∂ µΦ) − (∂ µΦ∗)Φ]

←→
= −ie Φ∗ ∂ µ Φ . (4.127)
One can add a potential term to L, of the form V (Φ∗Φ), without spoiling the
gauge invariance, so that we denote
Lmatter = Lfree − V (Φ∗Φ) . (4.128)
A minimal example is V = λ(Φ∗Φ)2 . With such a modification, the equations

of motion then read
∂V
Φ∗ + m2 Φ∗ = − , (4.129)
∂Φ
∂V
Φ + m2 Φ = − ∗ . (4.130)
∂Φ
Obviously, since Lmatter and Lfree differ only through powers of Φ∗Φ, the Noether
current (4.127) remains identical.
4.5.2. Pure photon part

As we have shown above, the Lagrangian which describes the dynamics of pho-
tons reads
1
Lem = − Fµν F µν . (4.131)
4
4.5.3. Interaction between matter and photons

Minimal Lagrangian through a dynamical current
We know that one can add a term −j µ Aµ , so that the equation of motion for Aµ
lead to the Maxwell equations. Doing, so, we immediately encounter a technical
67
4. Electrodynamics
problem: indeed, this is valid for jµ being external, non dynamical, but what if
j µ is promoted to be dynamical, being itself constructed from dynamical fields?
We already have a candidate to start with:

µ ←→
−jNoether Aµ = ie Φ∗ ∂ µ ΦAµ . (4.132)
But this new term in the Lagrangian, which now involves derivatives of the
fields Φ and Φ∗, will modify the Noether current itself, so that this additional
term should be itself modified, changing the current, etc. My gosh! Do we enter
an endless loop?
Let us show that one can find a minimal and consistent solution to this
problem. For that purpose, we write
L = Lem + Lmatter + Lint . (4.133)
Let us formulate the problem.
⋄ The Euler-Lagrange equations applied to L for Aµ

∂L ∂L
∂σ = (4.134)
∂(∂σ Aρ ) ∂Aρ
should lead to the Maxwell equations. Two remarks are then in order:
- Since the dynamics of Aµ (its time derivative, and by covariance any of its
derivatives) is inside Lem , only Lem contributes to the l.h.s. of Eq. (4.134).
- The r.h.s. of Eq. (4.134) gets contributions from Lint alone.
Thus, Eq. (4.134) actually reads

∂Lem ∂Lint
∂σ = , (4.135)
∂(∂σ Aρ ) ∂Aρ
or equivalently
∂Lint
−∂σ F σρ = = −Jρ (4.136)
∂Aρ
where Jρ is the full Noether current, since Maxwell’s equations with this
current should hold.
68
⋄ The full Noether current is given by

1 ∂L ∂L
Jµ = δΦ + ∗
δΦ∗ (4.137)
δα ∂(∂µΦ) ∂(∂µΦ )
so that

µ ∂Lmatter ∂Lmatter ∗
J = ie Φ− Φ
∂(∂µΦ) ∂(∂µΦ∗)

∂Lint ∂Lint ∗
+ie Φ− Φ , (4.138)
∂(∂µΦ) ∂(∂µΦ∗)
In this expression, the first part, coming from Lmatter , is the current
←→
j µ = −ie Φ∗ ∂ µ Φ , (4.139)
see Eq. (4.127), which is of course independent of Aµ .

Our problem is thus to look for Lint , solution of the coupled equations (4.136)
and (4.138). We are looking for a minimal solution (the problem has per se no
unique solution).
First, let us integrate Eq. (4.136) with respect to Aρ : separating the term
−j µ Aµ out of Lint , i.e.
←→
Lint = −j µ Aµ + L′int = ie Φ∗ ∂ µ ΦAµ + L′int . (4.140)
We now make a minimal assumption: we look for a solution with L′int indepen-
dent of ∂µ Φ and ∂µ Φ∗.
Thus,
∂Lint
ρ
= ieAρ Φ∗, (4.141)
∂(∂ Φ)
∂Lint
= −ieAρ Φ (4.142)
∂(∂ ρΦ∗)
and
Jρ = jρ + ie [ieAρ Φ∗ Φ − (−ie)AρΦΦ∗]
= jρ − 2e2Aρ Φ∗Φ . (4.143)
The last step is to use Eq. (4.136), which thus reads

∂Lint
ρ
= −Jρ = −jρ + 2e2Aρ Φ∗Φ (4.144)
∂A
69
4. Electrodynamics
and to solve it for Lint . We thus finally get
Lint = −j µ Aµ + e2 A2 Φ∗Φ . (4.145)
In conclusion, we have constructed the minimal Lagrangian
L = Lem + Lmatter + Lint (4.146)

1
= − Fµν F µν + (∂µ Φ) (∂ µ Φ)∗ − m2 ΦΦ∗ − V (Φ∗Φ)
4
←→
+ieAµ (Φ∗ ∂ µ Φ) + e2 A2 Φ∗Φ .
Let us summarize the equations of motion satisfied by the fields Aµ , Φ and

Φ∗ .
⋄ First, the equation of motion for Aµ reads
∂µ F µν = J µ (4.147)
with J µ given by Eq. (4.143).
⋄ Second, the equations of motion for Φ and Φ∗, using respectively

∂L ∂L
∂σ = (4.148)
∂(∂σ Φ) ∂Φ
and
∂L ∂L
∂σ = , (4.149)
∂(∂σ Φ∗ ) ∂Φ∗
read4
∂V
(∂µ + ieAµ ) (∂ µ + ieAµ ) + m2 Φ∗ = − . (4.150)
∂Φ
and
∂V
(∂µ − ieAµ ) (∂ µ − ieAµ ) + m2 Φ = − ∗ . (4.151)
∂Φ
4
Beware on the fact that in these two equations, the differential operators ∂µ and ∂ µ act on any structure
which is on their right (e.g. Aµ , Φ).
70
Conservation of the current J µ

Let us take a short breath and verify the consistency of what we have done
so far. As we know, the current J µ should be conserved. Let us check that
our constructed current indeed satisfies this constraint, a fact which is not
completely obvious from its explicit expression (4.143).
A direct computation gives
∂µ J µ = −ie(∂µΦ∗)(∂ µΦ) + ie(∂ µΦ)(∂µΦ∗) − ieΦ∗Φ + ie(Φ∗)Φ
−2e2(∂µ Aµ)Φ∗Φ − 2e2Aµ (∂µ Φ∗)Φ − 2e2 AµΦ∗ (∂µΦ) . (4.152)
The first two terms cancel each other. For the other ones, things are a lit-
tle bit more tricky. The two equations of motion (4.150) and (4.151) can be
respectively expanded as
∂V
( + m2 )Φ∗ + ie(∂µ Aµ )Φ∗ + 2ieAµ(∂µ Φ∗) − e2 A2Φ∗ = − (4.153)
∂Φ
and
∂V
( + m2 )Φ − ie(∂µAµ )Φ − 2ieAµ (∂µΦ) − e2 A2Φ = − . (4.154)
∂Φ∗
Computing ieΦ × (4.153) − ieΦ∗ × (4.154) thus gives
ieΦ(Φ∗) − ieΦ∗(Φ) − 2e2Aµ (∂µ Φ∗)Φ − 2e2Aµ Φ∗ (∂µΦ)

∂V ∗ ∂V
= −ie Φ −Φ =0 (4.155)
∂Φ ∂Φ∗
which explicitly shows, when inserted in Eq. (4.152), that the current J µ is
indeed conserved.
Minimal Lagrangian from gauge invariance

One may have noticed that the structure of the Lagrangian (4.146) is very
peculiar, as one may guess from the equations of motion (4.150) and (4.151).
Indeed, L can be rewritten has
1
L = − Fµν F µν + (DµΦ)∗(Dµ Φ) − m2 Φ∗ Φ − V (Φ∗Φ) (4.156)
4
with the covariant derivative defined as5
Dµ = ∂µ − ieAµ , (4.157)
5
The present convention is the same as the one used in Halzen-Martin. It is also consistent with Itzykson-Zuber
and Peskin-Schroeder. In these two last references, one should pay attention to the fact that e is the electric
charge of the electron, i.e. e = −|e|, while we use here the convention that e = |e|.
71
4. Electrodynamics
so that the equation of motion (4.150) and (4.151) read

∗ ∂V
D2 + m2 Φ∗ = − , (4.158)
∂Φ
2 ∂V
D + m2 Φ = − ∗ . (4.159)
∂Φ
Let us make a step back, and re-examine the problem of coupling matter to
photon, just starting from the Lagrangian Lmatter on which we impose gauge
invariance, as a general principle.
We thus consider U (1) local transformation, named gauge transformation:
Φ(x) → Φ(x) eieα(x) (4.160)

Φ∗(x) → Φ∗(x) e−ieα(x) . (4.161)
Thus,
∂µ Φ(x) → ∂µ Φ(x) eieα(x) + ie ∂µ α(x)Φ(x) eieα(x) (4.162)

∂µ Φ∗(x) → ∂µ Φ∗(x) eieα(x) − ie ∂µ α(x)Φ∗(x) e−ieα(x) . (4.163)
Clearly, while −m2 Φ∗Φ − V (Φ∗Φ) is gauge invariant, this invariance is broken
by the term (∂µΦ(x))∗∂ µ Φ(x).
The way to restore this gauge invariance is to introduce a new field Aµ , which
carries a µ index like ∂µ (it is thus a spin one field, as shown by the study of
the representations of the Lorentz group). These two are combined to built up
the covariant derivative
Dµ = ∂µ − ieAµ .
Suppose now that under a gauge transformation, Aµ transforms as
Aµ (x) → Aµ (x) + ∂ µ α(x) , (4.164)
which we already encountered in Eq. (4.76) when discussing gauge invariance

of Maxwell’s equations when expressed in terms of Aµ . One then immediately
see that
Dµ Φ(x) → [∂µ Φ(x) + ie∂µ α(x) Φ(x) − ieAµ (x) Φ(x) − ie∂µ α(x) Φ(x)] eieα(x)
= [∂µ Φ(x) − ieAµ (x) Φ(x)] eieα(x)
= [Dµ Φ(x)] eieα(x) (4.165)
and similarly
[DµΦ(x)]∗ → [Dµ Φ(x)]∗ e−ieα(x) . (4.166)
72
Therefore, making the minimal replacement
[∂µ Φ(x)]∗ [∂µ Φ(x)] → [Dµ Φ(x)]∗ [Dµ Φ(x)] , (4.167)
the kinetic part of the Lagrangian becomes gauge invariant. Adding this term
to the mass and potential terms for the fields Φ and Φ∗, and to the pure QED
Lagrangian, one therefore gets the full QED gauge invariant Lagrangian which
we obtained earlier, see Eq. (4.156).
This construction can be extended to other groups. This is the essence of
the Yang-Mills construction. This is THE way to construct a dynamical field
theory which couples matter and gauge fields.
For example, passing from U (1) to U (1) × SU (2) gauge invariance led to
the construction of the electroweak theory, with quarks and leptons (electron,
muon, tau and their associated neutrinos) as matter fields, carrying charges
under this group, and γ, W ±, Z 0 as gauge fields.
Similarly, having a gauge theory based on the (color) group SU (3) leads to the
(quantum) chromodynamics (QCD), the modern theory of strong interaction,
with quarks as matter fields, and gluons as gauge fields.
This leads to the Standard Model, based on the gauge group U (1) × SU (2) ×
SU (3).
One should note that the gauge field dynamics itself is governed by a term of
the type (4.131). In general, as it is the case for the Standard Model, the gauge
group is non-abelian. The consequence is that the gauge fields themselves carry
a charge: by construction, gauge fields live in the adjoint representation of the
group. For an abelian group, this representation is trivial, and indeed we know
that the photon as no charge, and therefore does not couple to itself. But for
a non-abelian group, the gauge fields acquire a charge, so that they can couple
to each other. This is the case of W ± which carry an electric charge, of Z 0 , W ±
which carry a weak isospin, of the gluons which carry a color charge. This is
technically hidden in the field strength F µν which is not anymore linear in the
gauge field.
We refer to lectures on group theory for more details.
73
5. From Lagrangians to
cross-sections
In this chapter, we will show how to compute a physical observable for a given
scattering process, relying on the Lagrangian built in the previous chapter. This
will be limited to the case of scalar matter, in order to avoid any complication
related to spin.
5.1. Perturbation theory

From the interaction part of the Lagrangian, since
L = T − U, (5.1)
we can identify
U = −Lint . (5.2)
In the scattering of a “spinless” electron off an electromagnetic potential Aµ ,

facing a transition between an initial state i to a final state f , using the covariant
notation of Chap. 3, the amplitude Tf i thus reads, at lowest order, according
to Eq. (3.30),
Z
Tf i = i hf |Lint |ii d4 x , (5.3)
where hf |Lint |ii, see Eq. (4.145), is given by

←→
hf |Lint |ii = ieAµ Φ∗f ∂ µ Φi + e2 A2Φ∗f Φi . (5.4)
From now on, we only consider the dominant e1 term, neglecting the e2 term,
so that the Lint as the simple structure of a current jµ interacting with the field
Aµ :
Lint = −jµ Aµ , (5.5)
75
5. From Lagrangians to cross-sections
where jµ is just the matter current

←→
jµ = −ieΦ∗ ∂ µ Φ . (5.6)
Let us consider now the matrix element jµf i = hf |jµ |ii of this current:
⋄ an incoming (spinless) electron of moment pi has a wave function
Φi(x) = Ni e−ipi·x (5.7)
⋄ an outgoing (spinless) electron of moment pf has a wave function
Φ∗f (x) = Nf e+ipf ·x . (5.8)
The normalization constants Ni and Nf in Eqs. (5.7) and (5.8) will be specified
later. We thus have
jµf i = −eNi Nf (pi + pf )µ ei(pf −pi)·x . (5.9)
The perturbation theory developed in Chap. 3 was time-dependent, but we

later on restricted ourselves to a time-independent potential. However, nothing
prevent us to take into account the time dependence of the field Aµ .
To understand how things are going, let us consider as an example the spinless
e µ scattering1, in the kinematics
− −
e− (pA )µ−(pB ) → e− (pC )µ− (pD ) (5.10)

(2)
as illustrated in Fig. 5.1. The current jµ is a source term for the Aµ , which
(1)
itself couples to the current jµ . To find this field Aµ , one should just solve the
Maxwell’s equation
(2) Aµ = jµ(2) (5.11)
for Aµ , with, using Eq. (5.9),
jµ(2) = −eNB ND (pB + pD )µ ei(pD −pB )·x . (5.12)
Thus, since
ei(pD −pB )·x = −(pD − pB )2 ei(pD −pB )·x , (5.13)

1
We do not consider here the little bit more involved situation of e− e− scattering, in which two contributions
would occur, because the two particle are indistinguishable).
76
5.1. Perturbation theory
(1)
jµ
pA pC
e− e −
γ ↓q
pB µ− µ− pD
(2)
jµ
Figure 5.1.: e− µ− scattering.
we get
1 (2)
Aµ = − j (5.14)
q2 µ
where q = pD − pB .
At lowest order, we get, after integration over x,
Z
1
Tf i = −i jµ(1) − 2 j (2) µ (x) d4x (5.15)
q
= iNA NB NC ND (2π)4δ (4) (pA + pB − pC − pD ) M (5.16)
where

g µν
µ
iM = [ie(pA + pC ) ] −i 2 [ie(pB + pD )ν ] (5.17)
q
(5.18)
vertex photon vertex (5.19)
propagator (5.20)
Note: q 6= 0, therefore the photon which propagates is virtual, or off-mass shell.
77
5.2. Cross-sections
5.2.1. normalization of a free particle wave-function
The wave-function of a particle of momentum p reads
Φ(x) = N e−ip·x . (5.21)
We have shown in Chap. 2 that for such a plane wave, ρ = 2E|N |2, see
Eq. (2.55).
One should note the consistency and the Lorentz invariance under boosts.
Indeed, ρ d3 x is a Lorentz invariant, since
Boost d3 x
d3 x −→ (5.22)
γ
Boost
2E −→ γ2E . (5.23)
In a given volume V , we normalize Φ such that there are 2E particles:

Z Z
ρ d V = 2E|N |2 d3 V = 2E
3
(5.24)
V
i.e.
1
N=√ . (5.25)
V
5.2.2. Transition rate per unit volume

Following Chap. 3, we introduce the transition rate per unit volume, for a
process A + B → C + D, as
|Tf i|2
Wf i = . (5.26)
TV
Using the same trick which we used in Chap. 3 to pass from Tf i to |Tf i|2 in
order to give a meaning to [δ (4) (pA + pB − pC − pD )]2, one δ (4) remains, and the
other (2π)4δ (4) should be replaced by T V . Finally, one gets
(4)
4δ (pA + pB − pC − pD )|M|2
Wf i = (2π) . (5.27)
V4
We define the concept of cross-section as
Wf i
cross-section = × (number of final states) . (5.28)
initial flux
78
5.2. Cross-sections
The reason why one divides Wf i by the initial flux, which is the density of
incoming states is that one wants to normalize the result independently of any
particular experimental setup (like the density of the target, of the beam).
5.2.3. Number of final states

In a volume V , with momenta inside d3p, there are
V d3 p
ρV (p) = states . (5.29)
(2π)3
This is a pure quantum mechanical effect. Consider first a particle in a one-
dimensional box of length L. The boundary conditions implies, since the wave-
function is of the form
Φ ∼ eipx x , (5.30)
that px L = n2π with n ∈ Z, and thus, passing from a sum over n to an

integration over px , the number of states in the range px to px + dpx is
dpx L
dn = , (5.31)
2π
from which the result (5.29) comes immediately after passing from the one-
dimensional to the three-dimensional case, denoting V = L3 .
The number of final states/particle is thus
ρV (p) V d3 p
= , (5.32)
2E (2π)32E
where we have normalized the wave-function such that there are 2E particles
in the volume V.
For outgoing particles C and D scattered into d3pC and d3 pD , we thus have:
V d3 pC V d3 pD
number of final states = . (5.33)
(2π)32EC (2π)32ED
5.2.4. Initial flux

In the lab frame, see Fig. 5.2, the number of A particles which go through the
surface of area L2 during t is
2EA 2
L |~vA |t . (5.34)
V
79
~vA
Figure 5.2.: Evaluation of the initial flux.
Thus, the number of beam particles passing through a unit area per unit of
time is
2EA
|~vA | . (5.35)
V
The number of target particles per unit volume is
2EB
. (5.36)
V
We thus finally define the initial flux as
2EA 2EB
|~vA | . (5.37)
V V
5.2.5. Cross-section
Putting all the factors together, we obtain
V2 1 2 4 (4) V d3 pC V d3 pD
dσ = |M| (2π) δ (p A + p B − p C − p D )
|~vA |2EA 2EB V 4 (2π)32EC (2π)32ED
(5.38)
so that the V dependence cancels, as expected. One can thus write the differ-
ential cross-section as
|M|2
dσ = d(P S) (5.39)
F
where the phase-space is given by
4 (4) V d3 pC V d3 pD
d(P S) = (2π) δ (pA + pB − pC − pD ) (5.40)
(2π)32EC (2π)32ED
80
5.2. Cross-sections
which is Lorentz invariant, see tutorial, and F is the incident flux,
F = |~vA |2EA 2EB . (5.41)
Exercise 5.1
For an arbitrary collision along a given axis, show that
F = |~vA − ~vB |2EA 2EB (5.42)

1/2
= 4 (pA · pB )2 − m2A m2B = 2K, (5.43)
where we have introduced the standard notation 2K frequently used in the

literature.
5.2.6. Cross-section in the center-of-mass system

A few technical steps are still needed before obtaining the expression of the
differential cross-section in a compact form.
Exercise 5.2
In the center-of-mass system, introducing
|~pA | = |~pA | = p∗i , (5.44)

|~pC | = |~pD | = p∗f , (5.45)
show that
F = 2K = 4p∗i W ∗ (5.46)
where
W ∗ = EA∗ + EB∗ (5.47)
is the total center-of-mass energy.
Exercise 5.3
Introducing the solid angle Ω for the particle C, such that2
d3 pC = p2f dpf d2Ω , (5.48)

2
As usual, d2 Ω = dϕ sin θdθ when expressed in terms of the usual spherical coordinates (θ, ϕ).
81
show that the phase-space can be expressed as
1 p∗f 2
d(P S) = 2 d Ω. (5.49)
4π 4W ∗
Finally, the differential cross-section can be written as

dσ 1 p∗f
2
= 2 ∗ |M|2 , (5.50)
d Ω c.m.s 64π s pi

where we have introduced the Mandelstam variable
s = (pA + pB )2 = W ∗2 . (5.51)
We refer to the tutorial for the explicit study of the e− e− → e− e− scattering.
82
Appendices
83
A. Elements of group theory
A.1. Group
A.1.1. Definitions
Définition 1.1 : Group
A group is a pair (G, ·) made of a set G and an operation acting on this set,
which associates two a pair of elements a and b of G an element a · b.
This law should satisfy four axioms:
Internal composition law: ∀a, b ∈ G, a · b ∈ G
Associativity: ∀a, b, c ∈ G, (a · b) · c = a · (b · c)
Existence of a neutral element: ∃ e ∈ G/ ∀a ∈ G, e · a = a · e = a.

The element e is called the neutral element of the group.
Existence of a symmetric: ∀a ∈ G, ∃ sym(a) such that

a · sym(a) = sym(a) · a = e.
Depending on the context, the group law · can be denoted using various symbols:
⋄ whenever the law is additive, it will be denoted as +, and the symmetric

element of a is named as its opposite, denoted as −a. The neutral element
is then denoted as 0.
Example: the (Z, +) group.
In practice, this is used only for abelian groups (see below).
⋄ whenever the law is multiplicative, it will be denoted as ×, and the sym-

metric element of a is named as its inverse, denoted as a−1 . The neutral
element is then denoted as 1.
Example: for n ∈ N∗ , the group made of the n-roots of unity, equipped
with ×, is a multiplicative group.
85
⋄ one will also encounter the notations ◦ and ∗.
Définition 1.2 : Abelian and non-abelian groups

If the group law · is commutative, the group is named abelian.
In the opposite case, the group is named non-abelian.
Définition 1.3 : Group order
The order of a group G is the cardinal of G.
⋄ If the cardinal is finite, G is said to be a finite group (!) and its order is
denoted as |G|.
⋄ When the group has an infinite order, the group can be discrete (topolog-
ically), or continuous.
A.1.2. Examples of groups

⋄ The trivial group G = ({0}, +), also denoted as 0, for an additive group
(or G = ({1}, ×), also denoted as 1, for a multiplicative group).
⋄ (Z, +) is a discrete abelian group.
⋄ U (1) = {eiθ /θ ∈ R}, the group of phases, is abelian.
⋄ If K is a field (examples: K = R, C), (K, +) and (K ⋆, ×) are abelian

groups.
⋄ For n ∈ N⋆ , the set of integers modulo n equipped with addition G =

(Z/nZ, +) is an abelian group, of order n.
⋄ G = (S(E), ◦) :
On a set E, the set S(E) of bijections from E to E, equipped with the
composition law ◦ for maps is a group.
⋄ Symmetric group Sn :
In the special case where E = {1, · · · , n}, one denotes Sn as the set S(E),
which is called the symmetric group of n elements. Its order is n!, and it
is non abelian for n > 3, made of permutations of the various n elements
{1, · · · , n} .
86
A.1. Group
⋄ Linear group GLn (K) :

The set of invertible n×n matrices with coefficient belonging to a field K,
equipped with matrix multiplication, is a group (non-abelian for n > 2).
⋄ Special Linear group SLn (K) :

The n × n matrices with coefficient belonging to a field K, equipped with
matrix multiplication, of determinant 1 (therefore invertible), is a group
(non-abelian for n > 2).
⋄ Orthogonal group O(n):

The set O(n) of n × n orthogonal matrices is the set of real matrices
satisfying O · Ot = Ot · O = 1, or equivalently, which leaves the bilinear
n
P
form (the scalar product) ~x · ~y = xi yi invariant, i.e. (O~x) · (O~y ) = ~x · ~y .
i=1
Equipped with matrix multiplication, it is non-abelian group for n > 2.
The defining relation of O(n) obviously imposes that det O = ±1.
⋄ Special orthogonal group SO(n):

Subset of matrices of O(n) of determinant +1.
⋄ Unitary group U (n):

The set U (n) of unitary matrices n × n is made of complex matrices
satisfying U U † = U † U = 1, or equivalently, which leaves the sesquilinear
P n
form (x, y) = x∗i yi invariant (this form is left-sesquilinear: linear with
i=1
respect to y, antilinear with respect to x, see bra-ket in quantum physics),
i.e. (U x, U y) = (x, y).
Equipped with matrix multiplication, it is a non-abelian group for n > 2.
The defining relation for U (n) imposes that | det U | = 1.
⋄ Special unitary group (SU (n), ·):

Subset of U (n) matrices of determinant +1.
⋄ Symplectic group Sp(2n, K) :

Consider a field K (typically R or C).
The set Sp(2n, K) of symplectic matrices 2n × 2n is made of matrices
with coefficients belonging to K, which leaves the antisymmetric form
87
xtg y invariant, where g is the antisymmetric matrix

0 1n
g=
−1n 0
i.e. satisfying the condition
S t gS = g .
A.2. Morphisms and subgroups

A.2.1. Maps, injections, surjections
Définition 1.4 : Function (or map)

A function f : E → F maps elements from E to elements of F : any element of
E as a single image in F
Définition 1.5 : Image of a function
The image of a function f : E → F is the set of images of elements of E,
denoted as f (E), or Im f :
f (E) = {f (x)/x ∈ E} .
Définition 1.6 : Direct image of a subset

More generally, given a function f : E → F , the image (or direct image) f (A)
of a subset A ⊂ E is the set of images of elements of A:
f (A) = {f (x)/x ∈ A} .
Définition 1.7 : Inverse image of a subset

Given a function f : E → F , the inverse image f −1(B) of a subset B ⊂ F is
the set of inverse image of elements of B:
f −1(B) = {x ∈ E/f (x) ∈ B} .
Définition 1.8 : Injective function (or injection)

The function f : E → F is injective, or one-to-one, if each element of F is
mapped to by at most one element of the domain:
88
∀x, x′ ∈ E, f (x) = f (x′) =⇒ x = x′ ,

or equivalently, if distinct elements of E map to distinct elements in F :
∀x, x′ ∈ E, x 6= x′ =⇒ f (x) 6= f (x′) .
Définition 1.9 : Surjective function (or surjection)

The function f : E → F is surjective, or onto, if each element of F is mapped
to by at least one element of E. That is, the image F (E) is equal to F :
∀y ∈ F, ∃x ∈ E /y = f (x) .
Définition 1.10 : Bijective function (or bijection)

The function f : E → F is bijective if each element of F is mapped to by
exactly one element of E. In other words, the function is both injective and
surjective:
∀y ∈ F, ∃! x ∈ E /y = f (x) .
Définition 1.11 : Inverse of a bijection

To any bijection f : E → F , one can associate the inverse function (which is
also a bijection) denoted as f −1, which maps any element y ∈ F to its unique
inverse image x ∈ E, i.e.
x = f −1(y) ⇔ y = f (x) .
Be aware that conventionally, the same notation f −1 is used for both the
inverse bijection and for the inverse image of a set.
A.2.2. Morphism
Définition 1.12 : Group morphism
Let (G, ⋆) et (G′ , ·) be two groups. A morphism (or homomorphism) from G to
G′ is a map f : G → G′ which satisfies
∀g1 , g2 ∈ G, f (g1 ⋆ g2 ) = f (g1) · f (g2) . (A.1)
In other words, it means that the map f preserves the group law on G and G′ .
Définition 1.13 : Group isomorphism
89
If furthermore f is bijective, it is easy to show that f −1 is also a group morphism.

Then, f is called an isomorphism. In other words, the image of the symmetric
of g ∈ G is the symmetric of f (g) in G′ .
The two groups G et G′ are said to be isomorph, a fact denoted as G ≃ G′ or
G∼ = G′ .
Définition 1.14 : Group automorphism
If G = G′ , then the group isomorphism f is called an automorphism.
Proposition 1.15 : Aut(G)
The set of automorphisms of a group G , equipped with the composition of
maps ◦, is a group denoted as Aut(G) .
Définition 1.16 : Kernel of a morphism
The kernel of a morphism f is
Ker (f ) = {g ∈ G/f (g) = e} . (A.2)
Définition 1.17 : Injective, surjective morphism
⋄ A morphism f is injective (or on-to-one) if and only if Ker f = e .
⋄ Besides, as we have seen above for the general case of a function, a mor-
phism f is surjective if and only if Im f = G′ .
⋄ A morphism is an isomorphism if and only if it is both injective and

surjective.
A.2.3. Subgroup
Définition 1.18 : Subgroup

Let H be a subset of G. A subgroup (H, ·) of a group (G, ·) is a group whose
law · is obtained by restricting the group law · on H × H.
Proposition 1.19 : For a subset H of G to be a subgroup of G, it is necessary

and sufficient that ∀a, b ∈ H, a · b−1 ∈ H.
Remark :
Note that the criterion 1.19 makes it possible to dispense with checking the
associativity, automatically satisfied by inclusion of H in G.
90
Examples :
⋄ Evidently, G and {e} are subgroups of G.
⋄ The intersection of an arbitrary family (finite or infinite) of subgroups of

a group G is a subgroup of G.
⋄ The union of two subgroups of G is a subgroup if an only if one of the

two is included in the other.
⋄ The subgroups of Z are the n Z for n ∈ N .
Propositions 1.20 :
⋄ The image and the kernel of a morphism f : G → G′ are subgroups of G′

and G respectively.
More generally:
⋄ The inverse image by f of any subgroup of G′ is a subgroup of G.
⋄ The image by f of any subgroup of G is a subgroup of G′ .
Examples :
⋄ Let K be a field. Then det : GLn (K) → K∗ is a morphism.

If E is a K−vector space of dimension n, then GLn (K, ·) ∼
= (GL(E), ◦),
the group of linear bijections from E to E.
The kernel of det is thus a subgroup of GLn (K), named the special linear
group, denoted as SLn (K) (it is thus the set of matrices of determinant 1,
with coefficients in K).
91
B. Levi-Civita symbols and tensors
B.1. Levi-Civita symbol in an euclidean space
B.1.1. 2d
In two dimensions, one denotes
εij = +1 if (i, j) = (1, 2)

= −1 if (i, j) = (2, 1)
= 0 if i = j . (B.1)
One readily checks that
εij εmn = δim δjn − δin δjm . (B.2)
After one contraction over a pair of indexes, one gets
εij εin = δjn , (B.3)
since
δiiδjn − δinδji = 2δjn − δjn = δjn . (B.4)
Starting from Eq. (B.3), one more contraction gives
ǫij ǫij = 2 . (B.5)
B.1.2. 3d
In three dimensions, one denotes
εijk = +1 if (i, j, k) = even permutation of (1, 2, 3)

= −1 if (i, j, k) = odd permutation of (1, 2, 3)
= 0 otherwise . (B.6)
93
One can show easily that

δiℓ δim δin

εijk εℓmn = δjℓ δjm δjn . (B.7)
δkℓ δkm δkn
Indeed, there are 6 terms, as can be seen from the fact in 3d, i, j and k, as well
as ℓ, m and n can take the 3 different values 1, 2, 3: thus i should be equal to
ℓ, m or n, which forces j to be one of the two remaining indexes among ℓ, m
or n, so that k is the last one, i.e. 3 × 2 × 1 = 6 choices. The first choice i = ℓ,
j = m, k = n, i.e δiℓδjm δkn, gives, without sum, ε2ijk = +1, while the 5 other
terms are simply obtained by permutation of indexes, accounting for the sign
provided by the product of the two ε.
One contraction gives
εijk εimn = δjm δkn − δjn δkm . (B.8)
One more contraction leads to
εijk εijn = 2δkn , (B.9)
since
δjj δkn − δjn δkj = 3δkn − δkn = 2δkn . (B.10)
Finally,
εijk εijk = 3! = 6 . (B.11)
B.1.3. General case

In dimension n, one denotes
εi1 i2 ···in = +1 if (i1, i2, · · · , in) = even permutation of (1, 2, · · · , n)

= −1 if (i1, i2, · · · , in) = odd permutation of (1, 2, · · · , n)
Consider a n × n matrix A, which has matrix elements aij . Its determinant

is given by
1
det A = εi1 ···in a1i1 · · · anin = εi ···i εj ···j ai j · · · ain jn (B.13)
n! 1 n 1 n 1 1
94
B.1. Levi-Civita symbol in an euclidean space
and
ai1 j1 · · · ain jn εi1 ···in = (det A) εj1···jn , (B.14)
which can be understood as the way the Levi-Civita symbol transforms under
an arbitrary linear transformation.
B.1.4. Transformation under O(n) and SO(n)

We recall that orthogonal transformations, elements of the group O(n),
Rn → Rn (B.15)
x 7→ O x with xi = Oi j xj (B.16)
are characterized by the constraint
O Ot = Ot O = 1 . (B.17)
Taking the determinant of both sides, one sees that det O = ±1 . The case
det O = 1 defines the subgroup SO(n), named special orthogonal group, also
named rotation group, by extension of the group of rotations in n = 2 dimension
or n = 3 dimension.
Inserting this inside the general transformation (B.14) of the Levi-Civita
symbol, we see that1
Oi1 j1 · · · Oin jn εi1 ···in = (det O) εj1 ···jn , with det O = ±1. (B.18)
The Levi-Civita symbol is thus a pseudo-tensor, since it naturally extends what

appends in 3d:
⋄ under a rotation, i.e. a special orthogonal transformation, of determinant

+1, the Levi-Civita is unchanged.
⋄ under an orthogonal transformation of determinant −1, for example a

reflection in an odd number of dimensions (like the familiar 3d space of
euclidean geometry), named parity P , it should acquire a minus sign if it
were a tensor, made for example as the direct product of the component
of a vector V i , as V i1 V i2 · · · V in . In the present case, it remains invariant,
thus the name of pseudo-tensor.
1
We are working here in the euclidean space Rn , with a trivial metric given by the identity matrix, therefore
the up or down position of indexes is arbitrary.
95
B.2. Levi-Civita tensor in 4d Minkowski space

We now consider the Minkowski space, with the usual metric gµν .
The 4-contravariant Levi-Civita tensor is defined as
εµνρσ = +1 if (µ, ν, ρ, σ) = even permutation of (0, 1, 2, 3)

= −1 if (µνρσ) = odd permutation of (0, 1, 2, 3)
Using as usual the tensor metric gµν and its inverse, one can descend or raise
any index.
Since det g = −1 , one has
εµνρσ = −εµνρσ . (B.20)
One should be careful with the fact that the normalization of εµνρσ is conven-
tional. In some textbook/articles, it is defined with an opposite sign. We use
here the convention of Jackson, Itzykson-Zuber and Peskin-Schroeder.
B.2.1. Contraction identities

The following successive contraction identities are satisfied.
′ ′ ′ ′ ′
εµνρσ εµ ν ρ σ = − det g αα α = µ, ν, ρ, σ row
α′ = µ′ , ν ′, ρ′, σ ′ column (B.21)
′ ′ ′ ′
εµνρσ εµν ρ σ = − det g αα α = ν, ρ, σ row
α′ = ν ′, ρ′ , σ ′ column (B.22)
′ ′
µνρσ ρ′ σ ′ ρρ σσ ρσ ′ ρ′ σ
ε εµν = −2 g g − g g . (B.23)
′ ′
εµνρσ εµνρ σ = −6g σσ . (B.24)
εµνρσ εµνρσ = −4! = −24 . (B.25)
96
B.2. Levi-Civita tensor in 4d Minkowski space
B.2.2. Relationship with the Levi-Civita symbol
ε0ijk = εijk = −ε0ijk . (B.26)
B.2.3. Behavior under Lorentz transformations

Under Lorentz transformations, the Levi-Civita tensor transforms as
εµνρσ 7→ Λµµ′ Λνν ′ Λρρ′ Λσσ′ εµ ν ρ σ = det Λ εµνρσ .

′ ′ ′ ′
(B.27)
Thus:
⋄ under proper transformations (L+ : det Λ = +1), εµνρσ is invariant. This

includes boosts and rotations.
⋄ under improper transformations (L− : det Λ = −1), εµνρσ is reversed.

This includes T and P separately.
In conclusion,
εµνρσ is a pseudo-tensor under Lorentz transformations. (B.28)
97

Manuscript M1 15 10 2021

Uploaded by

Manuscript M1 15 10 2021

Uploaded by

Master General Physics

Major Particles, Nuclei and Universe

Particles and Symmetries

2. The Klein-Gordon equation 25

2.4.4. Charge current and reinterpretation of the negative en-

II. Introduction to Quantum Electrodynamics 39

5. From Lagrangians to cross-sections 75

B. Levi-Civita symbols and tensors 93

1.1. Space-time interval

Example : Emission of a photon in π 0 → γγ disintegration:

The signal propagates at the speed of light (photon). It is detected at the

From the definition (1.1), the infinitesimal interval square reads

ds2 = c2 dt2 − dx2 − dy 2 − dz 2 . (1.2)

If ds2 = 0 in an inertial frame K, then ds′2 = 0 in any other inertial frame K ′.

Metric: it is given by the matrix

⋄ (∆s)2 > 0 time-like interval

⋄ (∆s)2 < 0 space-like interval

⋄ (∆s)2 = 0 light-like interval ,

as illustrated in Fig. 1.1.

The infinitesimal square of the space-time interval thus reads

ds2 = gµν dxµdxν . (1.4)

Figure 1.1.: The 3 different kinds of intervals.

A trajectory is given by a curve X µ (s) where s is an arbitrary parametrization

dt = γdτ with γ = √1 and β~ = ~vc . (1.9)

1.2. Poincaré and Lorentz groups

ds′ 2 = ds2 (1.10)

translation (4 real parameters)

The condition (1.10) reads

gαβ dx′αdx′β = gµν dxµ dxν ,

A(µρ)ν + A(νρ)µ = 0 , (1.13)

where A is symmetric with respect to indexes inside brackets.

µ ↔ ρ gives A(µρ)ν + A(µν)ρ = 0 , (1.14)

ν ↔ ρ gives A(µν)ρ + A(νρ)µ = 0 . (1.15)

We use matrix notations, namely

gαβ Λαµ Λβν = gµν . (1.16)

• algebraically, if Λ and Λ′ belong to L, then

and thus that Λ Λ′ ∈ L .

The set P of Poincaré transformations, characterized by a pair (a, Λ) where

given two Poincaré transformations (a, Λ) and (a′ , Λ′ ), it is an easy exercise to

(a, Λ) · (a′ , Λ′) = (Λ′ a + a′ , Λ′ Λ) (1.17)

of the required form (a′′ , Λ′′ ).

This is easily recovered in the following way: Λ being a linear transformation,

v ′ · v = (v ′)t gv = (Λv ′)t g(Λv) = (v ′)t Λt gΛv (1.19)

which is equivalent to (1.18).

1.2.2. Structure of the Lorentz group

The Lorentz group O(3, 1) has 4 connected component. Each of

Simply and multiply connected space:

A space is simply connected if any loop in this space can be contracted to a

Figure 1.2.: The four connected components of L.

These various connected components are related by discrete transformations:

• space inversion (or parity):

This is illustrated in Fig. 1.2.

L leaves the 3 regions x2 > 0, x2 = 0 (light-cone) and x2 < 0 invariant.

The restricted Lorentz group L↑+ is non compact.

1.2.3. The two types of Lorentz transformations

0 6 θ 6 π, 0 6 φ < 2π, 0 6 ψ 6 π , (1.20)

Show that a passive rotation of axis ~n (~n2 = 1) and angle Ψ reads

~x′ = (~x − ~n(~n · ~x)) cos θ − ~n ∧ ~x sin θ + ~n(~n · ~x) (1.22)

Indication: consider how each part of the decomposition of an arbitrary vector

~x = ~xk + ~x⊥ = [(~x · ~n)~n] + [~x − (~x · ~n)~n]

transforms under a rotation.

In the passive point of view, Λ encodes the transformation from a frame F to

One should note that introducing β(−1 6 β 6 1) as

Generally speaking, a boost is therefore characterized by three real numbers

1.2.4. Number of parameters

where R is a rotation and B is a boost. We thus deduce that

L depends on 6 real parameters. (1.32)

where A is an invertible n × n matrix, with entries Ai j . This reads explicitly3

A vector ~v can be expanded in the basis B, with notations which emphasize

~v = v i[B]~ei = B v[B] , (1.36)

where the column vector of components v i[B] is denoted as v[B],

ei (~ej ) = δji . (1.42)

Any linear form ℓ ∈ E ∗ can then be expanded in this basis as

ℓ = ℓi[B]ei = ℓ[~ei ]ei . (1.43)

ℓ(~ei ) = ℓj [B]ej (~ei) = ℓj [B]δji = ℓi [B] . (1.44)

This expansion can be written in a more compact and useful form, as